Search

Reverse thinking plays a crucial role in human reasoning. Humans can reason not only from a problem to a solution but also in reverse, i.e., start from the solution and reason towards the problem. This often enhances overall reasoning performance as it enables consistency checks between their forward and backward thinking. To enable Large Language Models (LLMs) to perform reverse thinking, we introduce Reverse-Enhanced Thinking (RevThink), a framework composed of data augmentation and learning objectives. In RevThink, we augment the dataset by collecting structured forward-backward reasoning from a teacher model, consisting of: (1) the original question, (2) forward reasoning, (3) backward question, and (4) backward reasoning. We then employ three objectives to train a smaller student model in a multi-task learning fashion: (a) generate forward reasoning from a question, (b) generate a backward question from a question, and (c) generate backward reasoning from the backward question. Experiments across 12 datasets covering commonsense, math, and logical reasoning show an average 13.53% improvement over the student model's zero-shot performance and a 6.84% improvement over the strongest knowledge distillation baselines. Moreover, our method demonstrates sample efficiency -- using only 10% of the correct forward reasoning from the training data, it outperforms a standard fine-tuning method trained on 10x more forward reasoning. RevThink also exhibits strong generalization to out-of-distribution held-out datasets., Comment: 20 pages

We present improved approximation algorithms for some problems in the related areas of Flexible Graph Connectivity and Capacitated Network Design. In the $(p,q)$-Flexible Graph Connectivity problem, denoted $(p,q)$-FGC, the input is a graph $G(V, E)$ where $E$ is partitioned into safe and unsafe edges, and the goal is to find a minimum cost set of edges $F$ such that the subgraph $G'(V, F)$ remains $p$-edge connected upon removal of any $q$ unsafe edges from $F$. In the related Cap-$k$-ECSS problem, we are given a graph $G(V,E)$ whose edges have arbitrary integer capacities, and the goal is to find a minimum cost subset of edges $F$ such that the graph $G'(V,F)$ is $k$-edge connected. We obtain a $7$-approximation algorithm for the $(1,q)$-FGC problem that improves upon the previous best $(q+1)$-approximation. We also give an $O(\log{k})$-approximation algorithm for the Cap-$k$-ECSS problem, improving upon the previous best $O(\log{n})$-approximation whenever $k = o(n)$. Both these results are obtained by using natural LP relaxations strengthened with the knapsack-cover inequalities, and then during the rounding process utilizing an $O(1)$-approximation algorithm for the problem of covering small cuts. We also show that the the problem of covering small cuts inherently arises in another variant of $(p,q)$-FGC. Specifically, we show $O(1)$-approximate reductions between the $(2,q)$-FGC problem and the 2-Cover$\;$Small$\;$Cuts problem where each small cut needs to be covered twice.

In this paper we study a combinatorial reconfiguration problem that involves finding an optimal sequence of swaps to move an initial configuration of tokens that are placed on the vertices of a graph to a final desired one. This problem arises as a crucial step in reducing the depth of a quantum circuit when compiling a quantum algorithm. We provide the first known constant factor approximation algorithms for the parallel token swapping problem on graph topologies that are commonly found in modern quantum computers, including cycle graphs, subdivided star graphs, and grid graphs. We also study the so-called stretch factor of a natural lower bound to the problem, which has been shown to be useful when designing heuristics for the qubit routing problem. Finally, we study the colored version of this reconfiguration problem where some tokens share the same color and are considered indistinguishable.

The data processing inequality is central to information theory and motivates the study of monotonic divergences. However, it is not clear operationally we need to consider all such divergences. We establish a simple method for Pinsker inequalities as well as general bounds in terms of $\chi^{2}$-divergences for twice-differentiable $f$-divergences. These tools imply new relations for input-dependent contraction coefficients. We use these relations to show for many $f$-divergences the rate of contraction of a time homogeneous Markov chain is characterized by the input-dependent contraction coefficient of the $\chi^{2}$-divergence. This is efficient to compute and the fastest it could converge for a class of divergences. We show similar ideas hold for mixing times. Moreover, we extend these results to the Petz $f$-divergences in quantum information theory, albeit without any guarantee of efficient computation. These tools may have applications in other settings where iterative data processing is relevant., Comment: Preliminary Version; Section IV presented at ITW 2024

Storytelling video generation (SVG) has recently emerged as a task to create long, multi-motion, multi-scene videos that consistently represent the story described in the input text script. SVG holds great potential for diverse content creation in media and entertainment; however, it also presents significant challenges: (1) objects must exhibit a range of fine-grained, complex motions, (2) multiple objects need to appear consistently across scenes, and (3) subjects may require multiple motions with seamless transitions within a single scene. To address these challenges, we propose DreamRunner, a novel story-to-video generation method: First, we structure the input script using a large language model (LLM) to facilitate both coarse-grained scene planning as well as fine-grained object-level layout and motion planning. Next, DreamRunner presents retrieval-augmented test-time adaptation to capture target motion priors for objects in each scene, supporting diverse motion customization based on retrieved videos, thus facilitating the generation of new videos with complex, scripted motions. Lastly, we propose a novel spatial-temporal region-based 3D attention and prior injection module SR3AI for fine-grained object-motion binding and frame-by-frame semantic control. We compare DreamRunner with various SVG baselines, demonstrating state-of-the-art performance in character consistency, text alignment, and smooth transitions. Additionally, DreamRunner exhibits strong fine-grained condition-following ability in compositional text-to-video generation, significantly outperforming baselines on T2V-ComBench. Finally, we validate DreamRunner's robust ability to generate multi-object interactions with qualitative examples., Comment: Project website: https://dreamrunner-story2video.github.io/

In a multi-tenant large language model (LLM) serving platform hosting diverse applications, some users may submit an excessive number of requests, causing the service to become unavailable to other users and creating unfairness. Existing fairness approaches do not account for variations in token lengths across applications and multiple LLM calls, making them unsuitable for such platforms. To address the fairness challenge, this paper analyzes millions of requests from thousands of users on MS CoPilot, a real-world multi-tenant LLM platform hosted by Microsoft. Our analysis confirms the inadequacy of existing methods and guides the development of FairServe, a system that ensures fair LLM access across diverse applications. FairServe proposes application-characteristic aware request throttling coupled with a weighted service counter based scheduling technique to curb abusive behavior and ensure fairness. Our experimental results on real-world traces demonstrate FairServe's superior performance compared to the state-of-the-art method in ensuring fairness. We are actively working on deploying our system in production, expecting to benefit millions of customers world-wide.

Recent text-to-video (T2V) diffusion models have demonstrated impressive generation capabilities across various domains. However, these models often generate videos that have misalignments with text prompts, especially when the prompts describe complex scenes with multiple objects and attributes. To address this, we introduce VideoRepair, a novel model-agnostic, training-free video refinement framework that automatically identifies fine-grained text-video misalignments and generates explicit spatial and textual feedback, enabling a T2V diffusion model to perform targeted, localized refinements. VideoRepair consists of four stages: In (1) video evaluation, we detect misalignments by generating fine-grained evaluation questions and answering those questions with MLLM. In (2) refinement planning, we identify accurately generated objects and then create localized prompts to refine other areas in the video. Next, in (3) region decomposition, we segment the correctly generated area using a combined grounding module. We regenerate the video by adjusting the misaligned regions while preserving the correct regions in (4) localized refinement. On two popular video generation benchmarks (EvalCrafter and T2V-CompBench), VideoRepair substantially outperforms recent baselines across various text-video alignment metrics. We provide a comprehensive analysis of VideoRepair components and qualitative examples., Comment: Project page: https://video-repair.github.io

We report measurements of the absolute branching fractions $\mathcal{B}(B_s^0 \to D_s^{\pm} X)$, $\mathcal{B}(B_s^0 \to D^0/\bar{D}^0 X)$, and $\mathcal{B}(B_s^0 \to D^{\pm} X)$, where the latter is measured for the first time. The results are based on a 121.4\,fb$^{-1}$ data sample collected at the $\Upsilon(10860)$ resonance by the Belle detector at the KEKB asymmetric-energy $e^+ e^-$ collider. We reconstruct one $B_s^0$ meson in $e^+e^- \to \Upsilon(10860) \to B_s^{*} \bar{B}_s^{*}$ events and measure yields of $D_s^+$, $D^0$, and $D^+$ mesons in the rest of the event. We obtain $\mathcal{B}(B_s^0 \to D_s^{\pm} X) = (68.6 \pm 7.2 \pm 4.0)\%$, $\mathcal{B}(B_s^0 \to D^0/\bar{D}^0 X) = (21.5 \pm 6.1 \pm 1.8)\%$, and $\mathcal{B}(B_s^0 \to D^{\pm} X) = (12.6 \pm 4.6 \pm 1.3)\%$, where the first uncertainty is statistical and the second is systematic. Averaging with previous Belle measurements gives $\mathcal{B}(B_s^0 \to D_s^{\pm} X) = (63.4 \pm 4.5 \pm 2.2)\%$ and $\mathcal{B}(B_s^0 \to D^0/\bar{D}^0 X) = (23.9 \pm 4.1 \pm 1.8)\%$. For the $B_s^0$ production fraction at the $\Upsilon(10860)$, we find $f_s = (21.4^{+1.5}_{-1.7})\%$., Comment: 23 pages, 9 figures, submitted to JHEP

The effect of columnar grain boundaries on the fracture toughness was investigated using micro-cantilever fracture testing with a bridge notch, and a unique hard coating consisting of two distinct microstructures: one with columnar grains and another with an epitaxial layer. The bridge-failure sequence qualitatively demonstrated the lower fracture toughness at the columnar-grained structure. Quantitatively, the load drops measured at bridge-failure also revealed a significant decrease in fracture toughness due to grain boundaries. Specifically, the fracture toughness decreased by around 30%, from 4.1 +/- 0.4 MPa m1/2 for epitaxial microstructure to 3.0 +/- 0.3 MPa m1/2 for columnar-grained structure. The fracture toughness of columnar-grained structure is 3.0 +/- 0.2 MPa m1/2 perpendicular to the growth direction higher than 2.7 +/- 0.1 MPa m1/2 along it. These findings suggest that future optimization of hard coatings should focus on grain boundary toughening, and the present toolbox proposes suitable techniques for such microstructure optimization., Comment: 4 figures

Daily fantasy baseball has shortened the life cycle of an entire fantasy season into a single day. As of today, it has become familiar with more than 10 million people around the world who participate in online fantasy. As daily fantasy continues to grow, the importance of selecting a winning lineup becomes more valuable. The purpose of this paper is to determine how accurate FanDuel current daily fantasy strategy of optimizing daily lineups are and utilize python and linear programming to build a lineup optimizer for daily fantasy sports with the goal of proposing a more accurate model to assist daily fantasy participants select a winning lineup.

Latest developments in Large Multimodal Models (LMMs) have broadened their capabilities to include video understanding. Specifically, Text-to-video (T2V) models have made significant progress in quality, comprehension, and duration, excelling at creating videos from simple textual prompts. Yet, they still frequently produce hallucinated content that clearly signals the video is AI-generated. We introduce ViBe: a large-scale Text-to-Video Benchmark of hallucinated videos from T2V models. We identify five major types of hallucination: Vanishing Subject, Numeric Variability, Temporal Dysmorphia, Omission Error, and Physical Incongruity. Using 10 open-source T2V models, we developed the first large-scale dataset of hallucinated videos, comprising 3,782 videos annotated by humans into these five categories. ViBe offers a unique resource for evaluating the reliability of T2V models and provides a foundation for improving hallucination detection and mitigation in video generation. We establish classification as a baseline and present various ensemble classifier configurations, with the TimeSFormer + CNN combination yielding the best performance, achieving 0.345 accuracy and 0.342 F1 score. This benchmark aims to drive the development of robust T2V models that produce videos more accurately aligned with input prompts.

We present measurements of $B \to K{}^{*}(892)\gamma$ decays using $365\,{\rm fb}^{-1}$ of data collected from 2019 to 2022 by the Belle~II experiment at the SuperKEKB asymmetric-energy $e^+e^-$ collider. The data sample contains $(387 \pm 6) \times 10^6$ $B\overline{B}$ events. We measure branching fractions ($\mathcal{B}$) and $C\!P$ asymmetries ($\mathcal{A}_{C\!P}$) for both $B^{0}\to K{}^{*0}\gamma$ and $B^{+}\to K{}^{*+}\gamma$ decays. The difference in $C\!P$ asymmetries ($\Delta \mathcal{A}_{C\!P}$) and the isospin asymmetry ($\Delta_{0+}$) between these neutral and charged channels are also measured. We obtain the following branching fractions and $C\!P$ asymmetries: $\mathcal{B} (B^{0} \to K{}^{*0}\gamma) = (4.14 \pm 0.10 \pm 0.11 ) \times 10^{-5}$, $\mathcal{B} (B^{+} \to K{}^{*+}\gamma) = (4.02 \pm 0.13 \pm 0.13 )\times 10^{-5}$, $\mathcal{A}_{C\!P} (B^{0} \to K{}^{*0}\gamma) = (-3.3 \pm 2.3 \pm 0.4 )\%$, and $\mathcal{A}_{C\!P} (B^{+} \to K{}^{*+}\gamma) = (-0.7 \pm 2.9 \pm 0.6 )\%$. The measured difference in $C\!P$ asymmetries is $\Delta \mathcal{A}_{C\!P} = (+2.6 \pm 3.8 \pm 0.7 )\%$, and the measured isospin asymmetry is $\Delta_{0+} = (+5.0 \pm 2.0 \pm 1.5 )\%$. The first uncertainties listed are statistical and the second are systematic. These results are consistent with world-average values and theory predictions.

In this paper, we introduce Motion-Grounded Video Reasoning, a new motion understanding task that requires generating visual answers (video segmentation masks) according to the input question, and hence needs implicit spatiotemporal reasoning and grounding. This task extends existing spatiotemporal grounding work focusing on explicit action/motion grounding, to a more general format by enabling implicit reasoning via questions. To facilitate the development of the new task, we collect a large-scale dataset called GROUNDMORE, which comprises 1,715 video clips, 249K object masks that are deliberately designed with 4 question types (Causal, Sequential, Counterfactual, and Descriptive) for benchmarking deep and comprehensive motion reasoning abilities. GROUNDMORE uniquely requires models to generate visual answers, providing a more concrete and visually interpretable response than plain texts. It evaluates models on both spatiotemporal grounding and reasoning, fostering to address complex challenges in motion-related video reasoning, temporal perception, and pixel-level understanding. Furthermore, we introduce a novel baseline model named Motion-Grounded Video Reasoning Assistant (MORA). MORA incorporates the multimodal reasoning ability from the Multimodal LLM, the pixel-level perception capability from the grounding model (SAM), and the temporal perception ability from a lightweight localization head. MORA achieves respectable performance on GROUNDMORE outperforming the best existing visual grounding baseline model by an average of 21.5% relatively. We hope this novel and challenging task will pave the way for future advancements in robust and general motion understanding via video reasoning segmentation

The difficulty of exploring and training online on real production systems limits the scope of real-time online data/feedback-driven decision making. The most feasible approach is to adopt offline reinforcement learning from limited trajectory samples. However, after deployment, such policies fail due to exogenous factors that temporarily or permanently disturb/alter the transition distribution of the assumed decision process structure induced by offline samples. This results in critical policy failures and generalization errors in sensitive domains like Real-Time Communication (RTC). We solve this crucial problem of identifying robust actions in presence of domain shifts due to unseen exogenous stochastic factors in the wild. As it is impossible to learn generalized offline policies within the support of offline data that are robust to these unseen exogenous disturbances, we propose a novel post-deployment shaping of policies (Streetwise), conditioned on real-time characterization of out-of-distribution sub-spaces. This leads to robust actions in bandwidth estimation (BWE) of network bottlenecks in RTC and in standard benchmarks. Our extensive experimental results on BWE and other standard offline RL benchmark environments demonstrate a significant improvement ($\approx$ 18% on some scenarios) in final returns wrt. end-user metrics over state-of-the-art baselines.

We advance the study of pure de Sitter supergravity by introducing a finite formulation of unimodular supergravity via the super-St\"uckelberg mechanism. Building on previous works, we construct a complete four-dimensional action of spontaneously broken ${\cal N}\!\!=\!\!1$ supergravity to all orders, which allows for de Sitter solutions. The introduction of finite supergravity transformations extends the super-St\"uckelberg procedure beyond the second order, offering a recursive solution to all orders in the goldstino sector. This work bridges the earlier perturbative approaches and the complete finite theory, opening new possibilities for de Sitter vacua in supergravity models and eventually string theory., Comment: 23 pages

Modern AI agents, driven by advances in large foundation models, promise to enhance our productivity and transform our lives by augmenting our knowledge and capabilities. To achieve this vision, AI agents must effectively plan, perform multi-step reasoning and actions, respond to novel observations, and recover from errors, to successfully complete complex tasks across a wide range of scenarios. In this work, we introduce Magentic-One, a high-performing open-source agentic system for solving such tasks. Magentic-One uses a multi-agent architecture where a lead agent, the Orchestrator, plans, tracks progress, and re-plans to recover from errors. Throughout task execution, the Orchestrator directs other specialized agents to perform tasks as needed, such as operating a web browser, navigating local files, or writing and executing Python code. We show that Magentic-One achieves statistically competitive performance to the state-of-the-art on three diverse and challenging agentic benchmarks: GAIA, AssistantBench, and WebArena. Magentic-One achieves these results without modification to core agent capabilities or to how they collaborate, demonstrating progress towards generalist agentic systems. Moreover, Magentic-One's modular design allows agents to be added or removed from the team without additional prompt tuning or training, easing development and making it extensible to future scenarios. We provide an open-source implementation of Magentic-One, and we include AutoGenBench, a standalone tool for agentic evaluation. AutoGenBench provides built-in controls for repetition and isolation to run agentic benchmarks in a rigorous and contained manner -- which is important when agents' actions have side-effects. Magentic-One, AutoGenBench and detailed empirical performance evaluations of Magentic-One, including ablations and error analysis are available at https://aka.ms/magentic-one

Document visual question answering (DocVQA) pipelines that answer questions from documents have broad applications. Existing methods focus on handling single-page documents with multi-modal language models (MLMs), or rely on text-based retrieval-augmented generation (RAG) that uses text extraction tools such as optical character recognition (OCR). However, there are difficulties in applying these methods in real-world scenarios: (a) questions often require information across different pages or documents, where MLMs cannot handle many long documents; (b) documents often have important information in visual elements such as figures, but text extraction tools ignore them. We introduce M3DocRAG, a novel multi-modal RAG framework that flexibly accommodates various document contexts (closed-domain and open-domain), question hops (single-hop and multi-hop), and evidence modalities (text, chart, figure, etc.). M3DocRAG finds relevant documents and answers questions using a multi-modal retriever and an MLM, so that it can efficiently handle single or many documents while preserving visual information. Since previous DocVQA datasets ask questions in the context of a specific document, we also present M3DocVQA, a new benchmark for evaluating open-domain DocVQA over 3,000+ PDF documents with 40,000+ pages. In three benchmarks (M3DocVQA/MMLongBench-Doc/MP-DocVQA), empirical results show that M3DocRAG with ColPali and Qwen2-VL 7B achieves superior performance than many strong baselines, including state-of-the-art performance in MP-DocVQA. We provide comprehensive analyses of different indexing, MLMs, and retrieval models. Lastly, we qualitatively show that M3DocRAG can successfully handle various scenarios, such as when relevant information exists across multiple pages and when answer evidence only exists in images., Comment: Project webpage: https://m3docrag.github.io

Self-alignment, whereby models learn to improve themselves without human annotation, is a rapidly growing research area. However, existing techniques often fail to improve complex reasoning tasks due to the difficulty of assigning correct rewards. An orthogonal approach that is known to improve correctness is self-consistency, a method applied at inference time based on multiple sampling in order to find the most consistent answer. In this work, we extend the self-consistency concept to help train models. We thus introduce self-consistency preference optimization (ScPO), which iteratively trains consistent answers to be preferred over inconsistent ones on unsupervised new problems. We show ScPO leads to large improvements over conventional reward model training on reasoning tasks such as GSM8K and MATH, closing the gap with supervised training with gold answers or preferences, and that combining ScPO with standard supervised learning improves results even further. On ZebraLogic, ScPO finetunes Llama-3 8B to be superior to Llama-3 70B, Gemma-2 27B, and Claude-3 Haiku., Comment: 16 pages, 3 figures

Contrastive learning attempts to learn representations from un-labeled data; it does so via a loss function that encourages the embedding of a point to be close to that of its augmentations, and far from the embeddings of random other points. This simple idea performs remarkably well, yet it is not precisely theoretically understood why this is the case. In this paper we analyze contrastive learning (specifically, the InfoNCE loss) in a natural context: dimensionality reduction in Gaussian Mixture Models. Crucially, we define an augmentation of a data point as being another independent draw from the same underlying mixture component. We show that vanilla InfoNCE is able to find the optimal lower-dimensional subspace even when the Gaussians are not isotropic -- something that vanilla spectral techniques cannot do. We further extend our analyses to multi-modal contrastive learning algorithms (e.g., CLIP). In this setting we show that contrastive learning learns the subset of fisher-optimal subspace, effectively filtering out all the noise from the learnt representations.

Contrastive decoding (CD) (Li et al., 2023) improves the next-token distribution of a large expert language model (LM) using a small amateur LM. Although CD is applied to various LMs and domains to enhance open-ended text generation, it is still unclear why CD often works well, when it could fail, and how we can make it better. To deepen our understanding of CD, we first theoretically prove that CD could be viewed as linearly extrapolating the next-token logits from a huge and hypothetical LM. We also highlight that the linear extrapolation could make CD unable to output the most obvious answers that have already been assigned high probabilities by the amateur LM. To overcome CD's limitation, we propose a new unsupervised decoding method called $\mathbf{A}$symptotic $\mathbf{P}$robability $\mathbf{D}$ecoding (APD). APD explicitly extrapolates the probability curves from the LMs of different sizes to infer the asymptotic probabilities from an infinitely large LM without inducing more inference costs than CD. In FactualityPrompts, an open-ended text generation benchmark, sampling using APD significantly boosts factuality in comparison to the CD sampling and its variants, and achieves state-of-the-art results for Pythia 6.9B and OPT 6.7B. Furthermore, in five commonsense QA datasets, APD is often significantly better than CD and achieves a similar effect of using a larger LLM. For example, the perplexity of APD on top of Pythia 6.9B is even lower than the perplexity of Pythia 12B in CommonsenseQA and LAMBADA., Comment: EMNLP 2024 Oral

This study presents an approach to developing a model for classifying abnormalities in video capsule endoscopy (VCE) frames. Given the challenges of data imbalance, we implemented a tiered augmentation strategy using the albumentations library to enhance minority class representation. Additionally, we addressed learning complexities by progressively structuring training tasks, allowing the model to differentiate between normal and abnormal cases and then gradually adding more specific classes based on data availability. Our pipeline, developed in PyTorch, employs a flexible architecture enabling seamless adjustments to classification complexity. We tested our approach using ResNet50 and a custom ViT-CNN hybrid model, with training conducted on the Kaggle platform. This work demonstrates a scalable approach to abnormality classification in VCE.

Active metamaterials are engineered structures that possess novel properties that can be changed after the point of manufacture. Their novel properties arise predominantly from their physical structure, as opposed to their chemical composition and can be changed through means such as direct energy addition into wave paths, or physically changing/morphing the structure in response to both a user or environmental input. Active metamaterials are currently of wide interest to the physics community and encompass a range of sub-domains in applied physics (e.g. photonic, microwave, acoustic, mechanical, etc.). They possess the potential to provide solutions that are more suitable to specific applications, or which allow novel properties to be produced which cannot be achieved with passive metamaterials, such as time-varying or gain enhancement effects. They have the potential to help solve some of the important current and future problems faced by the advancement of modern society, such as achieving net-zero, sustainability, healthcare and equality goals. Despite their huge potential, the added complexity of their design and operation, compared to passive metamaterials creates challenges to the advancement of the field, particularly beyond theoretical and lab-based experiments. This roadmap brings together experts in all types of active metamaterials and across a wide range of areas of applied physics. The objective is to provide an overview of the current state of the art and the associated current/future challenges, with the hope that the required advances identified create a roadmap for the future advancement and application of this field.

We measure the time-integrated CP asymmetry in $D^{0} \rightarrow K^{0}_{S}K^{0}_{S}$ decays reconstructed in $e^{+}e^{-} \rightarrow c\overline{c}$ events collected by the Belle and Belle II experiments. The corresponding data samples have integrated luminosities of 980 fb$^{-1}$ and 428 fb$^{-1}$, respectively. The $D^{0}$ decays are required to originate from the $D^{*+} \rightarrow D^{0}\pi^{+}$ decay, which determines the charm flavor at production time. A control sample of $D^{0} \rightarrow K^{+}K^{-}$ decays is used to correct for production and detection asymmetries. The result, $(-1.4\pm1.3{\rm(stat)}\pm0.1{\rm (syst)})\%$, is consistent with previous determinations and with CP symmetry., Comment: 10 pages, 3 figures. arXiv admin note: text overlap with arXiv:2410.22961

Large Language Models (LLMs) often exhibit positional bias in long-context settings, under-attending to information in the middle of inputs. We investigate the presence of this bias in long-form summarization, its impact on faithfulness, and various techniques to mitigate this bias. To consistently evaluate faithfulness, we first compile a benchmark of eight human-annotated long-form summarization datasets and perform a meta-evaluation of faithfulness metrics. We show that LLM-based faithfulness metrics, though effective with full-context inputs, remain sensitive to document order, indicating positional bias. Analyzing LLM-generated summaries across six datasets, we find a "U-shaped" trend in faithfulness, where LLMs faithfully summarize the beginning and end of documents but neglect middle content. Perturbing document order similarly reveals models are less faithful when important documents are placed in the middle of the input. We find that this behavior is partly due to shifting focus with context length: as context increases, summaries become less faithful, but beyond a certain length, faithfulness improves as the model focuses on the end. Finally, we experiment with different generation techniques to reduce positional bias and find that prompting techniques effectively direct model attention to specific positions, whereas more sophisticated approaches offer limited improvements. Our data and code are available in https://github.com/meetdavidwan/longformfact., Comment: 18 pages

We perform a model-independent measurement of the $D^0$-$\overline{D}{}^0$ mixing parameters using samples of $e^+e^-$-collision data collected by the Belle and Belle II experiments that have integrated luminosities of $951\ \text{fb}^{-1}$ and $408\ \text{fb}^{-1}$, respectively. Approximately $2.05\times10^6$ neutral $D$ mesons are reconstructed in the $D^0\rightarrow K^0_{S}\pi^+\pi^-$ channel, with the neutral $D$ flavor tagged by the charge of the pion in the $D^{*+}\rightarrow D^0\pi^+$ decay. Assuming charge-parity symmetry, the mixing parameters are measured to be $ x = (4.0\pm1.7\pm0.4)\times 10^{-3} $ and $ y = (2.9\pm1.4\pm0.3)\times 10^{-3}$, where the first uncertainties are statistical and the second systematic. The results are consistent with previous determinations.

Autonomous systems, such as self-driving cars and drones, have made significant strides in recent years by leveraging visual inputs and machine learning for decision-making and control. Despite their impressive performance, these vision-based controllers can make erroneous predictions when faced with novel or out-of-distribution inputs. Such errors can cascade into catastrophic system failures and compromise system safety. In this work, we compute Neural Reachable Tubes, which act as parameterized approximations of Backward Reachable Tubes to stress-test the vision-based controllers and mine their failure modes. The identified failures are then used to enhance the system safety through both offline and online methods. The online approach involves training a classifier as a run-time failure monitor to detect closed-loop, system-level failures, subsequently triggering a fallback controller that robustly handles these detected failures to preserve system safety. For the offline approach, we improve the original controller via incremental training using a carefully augmented failure dataset, resulting in a more robust controller that is resistant to the known failure modes. In either approach, the system is safeguarded against shortcomings that transcend the vision-based controller and pertain to the closed-loop safety of the overall system. We validate the proposed approaches on an autonomous aircraft taxiing task that involves using a vision-based controller to guide the aircraft towards the centerline of the runway. Our results show the efficacy of the proposed algorithms in identifying and handling system-level failures, outperforming methods that rely on controller prediction error or uncertainty quantification for identifying system failures.

Italian ryegrass is a grass weed commonly found in winter wheat fields that are competitive with winter wheat for moisture and nutrients. Ryegrass can cause substantial reductions in yield and grain quality if not properly controlled with the use of herbicides. To control the cost and environmental impact we detect weeds in drone and satellite imagery. Satellite imagery is too coarse to be used for precision spraying, but can aid in planning drone flights and treatments. Drone images on the other hand have sufficiently good resolution for precision spraying. However, ryegrass is hard to distinguish from the crop and annotation requires expert knowledge. We used the Python segmentation models library to test more than 600 different neural network architectures for weed segmentation in drone images and we map accuracy versus the cost of the model prediction for these. Our best system applies herbicides to over 99% of the weeds while only spraying an area 30% larger than the annotated weed area. These models yield large savings if the weed covers a small part of the field., Comment: 5 pages, 3 figures

The use of large language models (LLMs) as evaluators has garnered significant attention due to their potential to rival human-level evaluations in long-form response assessments. However, current LLM evaluators rely heavily on static, human-defined criteria, limiting their ability to generalize across diverse generative tasks and incorporate context-specific knowledge. In this paper, we propose a novel Self-Assessing LLM framework that integrates Context-Aware Criteria (SALC) with dynamic knowledge tailored to each evaluation instance. This instance-level knowledge enhances the LLM evaluator's performance by providing relevant and context-aware insights that pinpoint the important criteria specific to the current instance. Additionally, the proposed framework adapts seamlessly to various tasks without relying on predefined human criteria, offering a more flexible evaluation approach. Empirical evaluations demonstrate that our approach significantly outperforms existing baseline evaluation frameworks, yielding improvements on average 4.8% across a wide variety of datasets. Furthermore, by leveraging knowledge distillation techniques, we fine-tuned smaller language models for criteria generation and evaluation, achieving comparable or superior performance to larger models with much lower cost. Our method also exhibits a improvement in LC Win-Rate in AlpacaEval2 leaderboard up to a 12% when employed for preference data generation in Direct Preference Optimization (DPO), underscoring its efficacy as a robust and scalable evaluation framework.

In this work, we study the cosmological effects of a tower of warm dark matter states on the cosmic microwave background (CMB) and on large-scale structure (LSS). For concreteness, we consider the $N$naturalness model, which is a proposed mechanism to solve the Higgs hierarchy problem. In this framework, the sector of particles of the Standard Model is copied $N$ times where the Higgs mass-squared value is the only parameter that changes between sectors. The other sectors are similar to our own, except their particles are proportionally heavier and cooler compared to the Standard Model sector. Since each sector is extremely weakly coupled to other sectors, direct observations of the new particles are not expected. The addition of new photon-like species and new neutrino-like species, however, can be detected through the CMB and in LSS data. These additional neutrinos form a tower of states with increasing mass and decreasing temperature compared to the SM neutrinos. This tower causes a more gradual suppression of the matter power spectrum across different comoving wavenumbers than a single warm dark matter state would. We quantitatively explore these effects in the $N$naturalness model and compute the parameter space allowed by the Planck 2018, weak lensing, and Lyman-$\alpha$ datasets. Depending on the underlying parameters, Planck 2018 and weak lensing data can require $N$naturalness to be tuned at the $10\%$ level for Dirac neutrinos and at the $5\%$ level for Majorana neutrinos. The additional neutrino states are crucial in constraining the model, particularly through their suppression of the power spectrum at small scales. The inclusion of many warm dark matter species is computationally very challenging and we make detailed assessments of our approximations and comment on potential future improvements., Comment: 36 pages, 15 captioned figures, 4 tables

Neural posterior estimation (NPE), a simulation-based computational approach for Bayesian inference, has shown great success in situations where posteriors are intractable or likelihood functions are treated as "black boxes." Existing NPE methods typically rely on normalizing flows, which transform a base distributions into a complex posterior by composing many simple, invertible transformations. But flow-based models, while state of the art for NPE, are known to suffer from several limitations, including training instability and sharp trade-offs between representational power and computational cost. In this work, we demonstrate the effectiveness of conditional diffusions as an alternative to normalizing flows for NPE. Conditional diffusions address many of the challenges faced by flow-based methods. Our results show that, across a highly varied suite of benchmarking problems for NPE architectures, diffusions offer improved stability, superior accuracy, and faster training times, even with simpler, shallower models. These gains persist across a variety of different encoder or "summary network" architectures, as well as in situations where no summary network is required. The code will be publicly available at \url{https://github.com/TianyuCodings/cDiff}.

We introduce the concept of a generative infinite game, a video game that transcends the traditional boundaries of finite, hard-coded systems by using generative models. Inspired by James P. Carse's distinction between finite and infinite games, we leverage recent advances in generative AI to create Unbounded: a game of character life simulation that is fully encapsulated in generative models. Specifically, Unbounded draws inspiration from sandbox life simulations and allows you to interact with your autonomous virtual character in a virtual world by feeding, playing with and guiding it - with open-ended mechanics generated by an LLM, some of which can be emergent. In order to develop Unbounded, we propose technical innovations in both the LLM and visual generation domains. Specifically, we present: (1) a specialized, distilled large language model (LLM) that dynamically generates game mechanics, narratives, and character interactions in real-time, and (2) a new dynamic regional image prompt Adapter (IP-Adapter) for vision models that ensures consistent yet flexible visual generation of a character across multiple environments. We evaluate our system through both qualitative and quantitative analysis, showing significant improvements in character life simulation, user instruction following, narrative coherence, and visual consistency for both characters and the environments compared to traditional related approaches., Comment: Project page: https://generative-infinite-game.github.io/

Four-dimensional scanning transmission electron microscopy (4D-STEM) is a powerful tool that allows for the simultaneous acquisition of spatial and diffraction information, driven by recent advancements in direct electron detector technology. Although 4D-STEM has been predominantly developed for and used in conventional TEM and STEM, efforts are being made to implement the technique in scanning electron microscopy (SEM). In this paper, we push the boundaries of 4D-STEM in SEM and extend its capabilities in three key aspects: (1) faster acquisition rate with reduced data size, (2) higher angular resolution, and (3) application to various materials including conventional alloys and focused ion beam (FIB) lamella. Specifically, operating the MiniPIX Timepix3 detector in the event-driven mode significantly improves the acquisition rate by a factor of a few tenths compared to conventional frame-based mode, thereby opening up possibilities for integrating 4D-STEM into various in situ SEM testing. Furthermore, with a novel stage-detector geometry, a camera length of 160 mm is achieved which improves the angular resolution amplifying its utility, for example, magnetic or electric field imaging. Lastly, we successfully imaged a nanostructured platinum-copper thin film with a grain size of 16 nm and a thickness of 20 nm, and identified annealing twins in FIB-prepared polycrystalline copper using virtual darkfield imaging and orientation mapping. This work demonstrates the potential of synergetic combination of 4D-STEM with in situ experiments, and broadening its applications across a wide range of materials.

Renal tumors, especially renal cell carcinoma (RCC), show significant heterogeneity, posing challenges for diagnosis using radiology images such as MRI, echocardiograms, and CT scans. U-Net based deep learning techniques are emerging as a promising approach for automated medical image segmentation for minimally invasive diagnosis of renal tumors. However, current techniques need further improvements in accuracy to become clinically useful to radiologists. In this study, we present an improved U-Net based model for end-to-end automated semantic segmentation of CT scan images to identify renal tumors. The model uses residual connections across convolution layers, integrates a multi-layer feature fusion (MFF) and cross-channel attention (CCA) within encoder blocks, and incorporates skip connections augmented with additional information derived using MFF and CCA. We evaluated our model on the KiTS19 dataset, which contains data from 210 patients. For kidney segmentation, our model achieves a Dice Similarity Coefficient (DSC) of 0.97 and a Jaccard index (JI) of 0.95. For renal tumor segmentation, our model achieves a DSC of 0.96 and a JI of 0.91. Based on a comparison of available DSC scores, our model outperforms the current leading models., Comment: 8 pages

Diffusion models have emerged as a powerful tool for image generation and denoising. Typically, generative models learn a trajectory between the starting noise distribution and the target data distribution. Recently Liu et al. (2023b) designed a novel alternative generative model Rectified Flow (RF), which aims to learn straight flow trajectories from noise to data using a sequence of convex optimization problems with close ties to optimal transport. If the trajectory is curved, one must use many Euler discretization steps or novel strategies, such as exponential integrators, to achieve a satisfactory generation quality. In contrast, RF has been shown to theoretically straighten the trajectory through successive rectifications, reducing the number of function evaluations (NFEs) while sampling. It has also been shown empirically that RF may improve the straightness in two rectifications if one can solve the underlying optimization problem within a sufficiently small error. In this paper, we make two key theoretical contributions: 1) we provide the first theoretical analysis of the Wasserstein distance between the sampling distribution of RF and the target distribution. Our error rate is characterized by the number of discretization steps and a new formulation of straightness stronger than that in the original work. 2) under a mild regularity assumption, we show that for a rectified flow from a Gaussian to any general target distribution with finite first moment (e.g. mixture of Gaussians), two rectifications are sufficient to achieve a straight flow, which is in line with the previous empirical findings. Additionally, we also present empirical results on both simulated and real datasets to validate our theoretical findings.

Large language models (LLMs) are susceptible to persuasion, which can pose risks when models are faced with an adversarial interlocutor. We take a first step towards defending models against persuasion while also arguing that defense against adversarial (i.e. negative) persuasion is only half of the equation: models should also be able to accept beneficial (i.e. positive) persuasion to improve their answers. We show that optimizing models for only one side results in poor performance on the other. In order to balance positive and negative persuasion, we introduce Persuasion-Balanced Training (or PBT), which leverages multi-agent recursive dialogue trees to create data and trains models via preference optimization to accept persuasion when appropriate. PBT consistently improves resistance to misinformation and resilience to being challenged while also resulting in the best overall performance on holistic data containing both positive and negative persuasion. Crucially, we show that PBT models are better teammates in multi-agent debates. We find that without PBT, pairs of stronger and weaker models have unstable performance, with the order in which the models present their answers determining whether the team obtains the stronger or weaker model's performance. PBT leads to better and more stable results and less order dependence, with the stronger model consistently pulling the weaker one up., Comment: Code: https://github.com/esteng/persuasion_balanced_training

Recent advances in diffusion models have significantly enhanced their ability to generate high-quality images and videos, but they have also increased the risk of producing unsafe content. Existing unlearning/editing-based methods for safe generation remove harmful concepts from models but face several challenges: (1) They cannot instantly remove harmful concepts without training. (2) Their safe generation capabilities depend on collected training data. (3) They alter model weights, risking degradation in quality for content unrelated to toxic concepts. To address these, we propose SAFREE, a novel, training-free approach for safe T2I and T2V, that does not alter the model's weights. Specifically, we detect a subspace corresponding to a set of toxic concepts in the text embedding space and steer prompt embeddings away from this subspace, thereby filtering out harmful content while preserving intended semantics. To balance the trade-off between filtering toxicity and preserving safe concepts, SAFREE incorporates a novel self-validating filtering mechanism that dynamically adjusts the denoising steps when applying the filtered embeddings. Additionally, we incorporate adaptive re-attention mechanisms within the diffusion latent space to selectively diminish the influence of features related to toxic concepts at the pixel level. In the end, SAFREE ensures coherent safety checking, preserving the fidelity, quality, and safety of the output. SAFREE achieves SOTA performance in suppressing unsafe content in T2I generation compared to training-free baselines and effectively filters targeted concepts while maintaining high-quality images. It also shows competitive results against training-based methods. We extend SAFREE to various T2I backbones and T2V tasks, showcasing its flexibility and generalization. SAFREE provides a robust and adaptable safeguard for ensuring safe visual generation., Comment: The first two authors contributed equally; Project page: https://safree-safe-t2i-t2v.github.io/

Recent self-rewarding large language models (LLM) have successfully applied LLM-as-a-Judge to iteratively improve the alignment performance without the need of human annotations for preference data. These methods commonly utilize the same LLM to act as both the policy model (which generates responses) and the reward model (which scores and ranks those responses). The ranked responses are then used as preference pairs to train the LLM via direct alignment technologies (e.g. DPO). However, it is noteworthy that throughout this process, there is no guarantee of accuracy in the rewarding and ranking, which is critical for ensuring accurate rewards and high-quality preference data. Empirical results from relatively small LLMs (e.g., 7B parameters) also indicate that improvements from self-rewarding may diminish after several iterations in certain situations, which we hypothesize is due to accumulated bias in the reward system. This bias can lead to unreliable preference data for training the LLM. To address this issue, we first formulate and analyze the generalized iterative preference fine-tuning framework for self-rewarding language model. We then introduce the regularization to this generalized framework to mitigate the overconfident preference labeling in the self-rewarding process. Based on this theoretical insight, we propose a Consistency Regularized sElf-rewarding lAnguage Model (CREAM) that leverages the rewarding consistency across different iterations to regularize the self-rewarding training, helping the model to learn from more reliable preference data. With this explicit regularization, our empirical results demonstrate the superiority of CREAM in improving both reward consistency and alignment performance. The code is publicly available at https://github.com/Raibows/CREAM.

Supervised Learning is a way of developing Artificial Intelligence systems in which a computer algorithm is trained on labeled data inputs. Effectiveness of a Supervised Learning algorithm is determined by its performance on a given dataset for a particular problem. In case of Supervised Learning problems, Stacking Ensembles usually perform better than individual classifiers due to their generalization ability. Stacking Ensembles combine predictions from multiple Machine Learning algorithms to make final predictions. Inspite of Stacking Ensembles superior performance, the overhead of Stacking Ensembles such as high cost, resources, time, and lack of explainability create challenges in real-life applications. This paper shows how we can strike a balance between performance, time, and resource constraints. Another goal of this research is to make Ensembles more explainable and intelligible using the Human-Centered approach. To achieve the aforementioned goals, we proposed a Human-Centered Behavior-inspired algorithm that streamlines the Ensemble Learning process while also reducing time, cost, and resource overhead, resulting in the superior performance of Supervised Learning in real-world applications. To demonstrate the effectiveness of our method, we perform our experiments on nine real-world datasets. Experimental results reveal that the proposed method satisfies our goals and outperforms the existing methods.

The widespread dissemination of false information through manipulative tactics that combine deceptive text and images threatens the integrity of reliable sources of information. While there has been research on detecting fake news in high resource languages using multimodal approaches, methods for low resource Indic languages primarily rely on textual analysis. This difference highlights the need for robust methods that specifically address multimodal fake news in Indic languages, where the lack of extensive datasets and tools presents a significant obstacle to progress. To this end, we introduce the Multimodal Multilingual dataset for Indic Fake News Detection (MMIFND). This meticulously curated dataset consists of 28,085 instances distributed across Hindi, Bengali, Marathi, Malayalam, Tamil, Gujarati and Punjabi. We further propose the Multimodal Multilingual Caption-aware framework for Fake News Detection (MMCFND). MMCFND utilizes pre-trained unimodal encoders and pairwise encoders from a foundational model that aligns vision and language, allowing for extracting deep representations from visual and textual components of news articles. The multimodal fusion encoder in the foundational model integrates text and image representations derived from its pairwise encoders to generate a comprehensive cross modal representation. Furthermore, we generate descriptive image captions that provide additional context to detect inconsistencies and manipulations. The retrieved features are then fused and fed into a classifier to determine the authenticity of news articles. The curated dataset can potentially accelerate research and development in low resource environments significantly. Thorough experimentation on MMIFND demonstrates that our proposed framework outperforms established methods for extracting relevant fake news detection features.

Due to the growing volume of user generated content, hashtags are employed as topic indicators to manage content efficiently on social media platforms. However, finding these vital topics is challenging in microvideos since they contain substantial information in a short duration. Existing methods that recommend hashtags for microvideos primarily focus on content and personalization while disregarding relatedness among users. Moreover, the cold start user issue prevails in hashtag recommendation systems. Considering the above, we propose a hybrid filtering based MIcro-video haSHtag recommendatiON MISHON technique to recommend hashtags for micro-videos. Besides content based filtering, we employ user-based collaborative filtering to enhance recommendations. Since hashtags reflect users topical interests, we find similar users based on historical tagging behavior to model user relatedness. We employ a graph-based deep neural network to model user to user, modality to modality, and user to modality interactions. We then use refined modality specific and user representations to recommend pertinent hashtags for microvideos. The empirical results on three real world datasets demonstrate that MISHON attains a comparative enhancement of 3.6, 2.8, and 6.5 reported in percentage concerning the F1 score, respectively. Since cold start users exist whose historical tagging information is unavailable, we also propose a content and social influence based technique to model the relatedness of cold start users with influential users. The proposed solution shows a relative improvement of 15.8 percent in the F1 score over its content only counterpart. These results show that the proposed framework mitigates the cold start user problem.

Visual instruction datasets from various distributors are released at different times and often contain a significant number of semantically redundant text-image pairs, depending on their task compositions (i.e., skills) or reference sources. This redundancy greatly limits the efficient deployment of lifelong adaptable multimodal large language models, hindering their ability to refine existing skills and acquire new competencies over time. To address this, we reframe the problem of Lifelong Instruction Tuning (LiIT) via data selection, where the model automatically selects beneficial samples to learn from earlier and new datasets based on the current state of acquired knowledge in the model. Based on empirical analyses that show that selecting the best data subset using a static importance measure is often ineffective for multi-task datasets with evolving distributions, we propose Adapt-$\infty$, a new multi-way and adaptive data selection approach that dynamically balances sample efficiency and effectiveness during LiIT. We construct pseudo-skill clusters by grouping gradient-based sample vectors. Next, we select the best-performing data selector for each skill cluster from a pool of selector experts, including our newly proposed scoring function, Image Grounding score. This data selector samples a subset of the most important samples from each skill cluster for training. To prevent the continuous increase in the size of the dataset pool during LiIT, which would result in excessive computation, we further introduce a cluster-wise permanent data pruning strategy to remove the most semantically redundant samples from each cluster, keeping computational requirements manageable. Training with samples selected by Adapt-$\infty$ alleviates catastrophic forgetting, especially for rare tasks, and promotes forward transfer across the continuum using only a fraction of the original datasets., Comment: First two authors contributed equally. Code: https://github.com/adymaharana/adapt-inf

Swift and accurate blood smear analysis is an effective diagnostic method for leukemia and other hematological malignancies. However, manual leukocyte count and morphological evaluation using a microscope is time-consuming and prone to errors. Conventional image processing methods also exhibit limitations in differentiating cells due to the visual similarity between malignant and benign cell morphology. This limitation is further compounded by the skewed training data that hinders the extraction of reliable and pertinent features. In response to these challenges, we propose an optimized Coupled Transformer Convolutional Network (CoTCoNet) framework for the classification of leukemia, which employs a well-designed transformer integrated with a deep convolutional network to effectively capture comprehensive global features and scalable spatial patterns, enabling the identification of complex and large-scale hematological features. Further, the framework incorporates a graph-based feature reconstruction module to reveal the hidden or unobserved hard-to-see biological features of leukocyte cells and employs a Population-based Meta-Heuristic Algorithm for feature selection and optimization. To mitigate data imbalance issues, we employ a synthetic leukocyte generator. In the evaluation phase, we initially assess CoTCoNet on a dataset containing 16,982 annotated cells, and it achieves remarkable accuracy and F1-Score rates of 0.9894 and 0.9893, respectively. To broaden the generalizability of our model, we evaluate it across four publicly available diverse datasets, which include the aforementioned dataset. This evaluation demonstrates that our method outperforms current state-of-the-art approaches. We also incorporate an explainability approach in the form of feature visualization closely aligned with cell annotations to provide a deeper understanding of the framework.

We present a measurement of the branching fraction and time-dependent charge-parity ($CP$) decay-rate asymmetries in $B^0 \to J/\psi \pi^0$ decays. The data sample was collected with the Belle~II detector at the SuperKEKB asymmetric $e^+e^-$ collider in 2019-2022 and contains $(387\pm 6)\times 10^6$ $B\overline{B}$ meson pairs from $\Upsilon(4S)$ decays. We reconstruct $392\pm 24$ signal decays and fit the $CP$ parameters from the distribution of the proper-decay-time difference of the two $B$ mesons. We measure the branching fraction to be $B(B^0 \to J/\psi \pi^0)=(2.02 \pm 0.12 \pm 0.10)\times 10^{-5}$ and the direct and mixing-induced $CP$ asymmetries to be $C_{CP}=0.13 \pm 0.12 \pm 0.03$ and $S_{CP}=-0.88 \pm 0.17 \pm 0.03$, respectively, where the first uncertainties are statistical and the second are systematic. We observe mixing-induced $CP$ violation with a significance of $5.0$ standard deviations for the first time in this mode.

This work tackles the challenge of efficiently selecting high-performance pre-trained vision backbones for specific target tasks. Although exhaustive search within a finite set of backbones can solve this problem, it becomes impractical for large datasets and backbone pools. To address this, we introduce Vision Backbone Efficient Selection (VIBES), which aims to quickly find well-suited backbones, potentially trading off optimality for efficiency. We propose several simple yet effective heuristics to address VIBES and evaluate them across four diverse computer vision datasets. Our results show that these approaches can identify backbones that outperform those selected from generic benchmarks, even within a limited search budget of one hour on a single GPU. We reckon VIBES marks a paradigm shift from benchmarks to task-specific optimization., Comment: 9 pages, 4 figures, under review at WACV 2025

Hamilton-Jacobi (HJ) reachability analysis is a powerful framework for ensuring safety and performance in autonomous systems. However, existing methods typically rely on a white-box dynamics model of the system, limiting their applicability in many practical robotics scenarios where only a black-box model of the system is available. In this work, we propose a novel reachability method to compute reachable sets and safe controllers for black-box dynamical systems. Our approach efficiently approximates the Hamiltonian function using samples from the black-box dynamics. This Hamiltonian is then used to solve the HJ Partial Differential Equation (PDE), providing the reachable set of the system. The proposed method can be applied to general nonlinear systems and can be seamlessly integrated with existing reachability toolboxes for white-box systems to extend their use to black-box systems. Through simulation studies on a black-box slip-wheel car and a quadruped robot, we demonstrate the effectiveness of our approach in accurately obtaining the reachable sets for blackbox dynamical systems.

There has been an increasing interest in detecting hallucinations in model-generated texts, both manually and automatically, at varying levels of granularity. However, most existing methods fail to precisely pinpoint the errors. In this work, we introduce QASemConsistency, a new formalism for localizing factual inconsistencies in attributable text generation, at a fine-grained level. Drawing inspiration from Neo-Davidsonian formal semantics, we propose decomposing the generated text into minimal predicate-argument level propositions, expressed as simple question-answer (QA) pairs, and assess whether each individual QA pair is supported by a trusted reference text. As each QA pair corresponds to a single semantic relation between a predicate and an argument, QASemConsistency effectively localizes the unsupported information. We first demonstrate the effectiveness of the QASemConsistency methodology for human annotation, by collecting crowdsourced annotations of granular consistency errors, while achieving a substantial inter-annotator agreement ($\kappa > 0.7)$. Then, we implement several methods for automatically detecting localized factual inconsistencies, with both supervised entailment models and open-source LLMs.

The availability of performant pre-trained models has led to a proliferation of fine-tuned expert models that are specialized to particular domains. This has enabled the creation of powerful and adaptive routing-based "Model MoErging" methods with the goal of using expert modules to create an aggregate system with improved performance or generalization. However, existing MoErging methods often prioritize generalization to unseen tasks at the expense of performance on held-in tasks, which limits its practical applicability in real-world deployment scenarios. We observe that current token-level routing mechanisms neglect the global semantic context of the input task. This token-wise independence hinders effective expert selection for held-in tasks, as routing decisions fail to incorporate the semantic properties of the task. To address this, we propose, Global and Local Instruction Driven Expert Router (GLIDER) that integrates a multi-scale routing mechanism, encompassing a semantic global router and a learned local router. The global router leverages LLM's advanced reasoning capabilities for semantic-related contexts to enhance expert selection. Given the input query and LLM, the router generates semantic task instructions that guide the retrieval of the most relevant experts across all layers. This global guidance is complemented by a local router that facilitates token-level routing decisions within each module, enabling finer control and enhanced performance on unseen tasks. Our experiments using T5-based models for T0 and FLAN tasks demonstrate that GLIDER achieves substantially improved held-in performance while maintaining strong generalization on held-out tasks. We also perform ablations experiments to dive deeper into the components of GLIDER. Our experiments highlight the importance of our multi-scale routing that leverages LLM-driven semantic reasoning for MoErging methods., Comment: Our code is available at https://github.com/UNITES-Lab/glider

Instruction following is a key capability for LLMs. However, recent studies have shown that LLMs often struggle with instructions containing multiple constraints (e.g. a request to create a social media post "in a funny tone" with "no hashtag"). Despite this, most evaluations focus solely on synthetic data. To address this, we introduce RealInstruct, the first benchmark designed to evaluate LLMs' ability to follow real-world multi-constrained instructions by leveraging queries real users asked AI assistants. We also investigate model-based evaluation as a cost-effective alternative to human annotation for this task. Our findings reveal that even the proprietary GPT-4 model fails to meet at least one constraint on over 21% of instructions, highlighting the limitations of state-of-the-art models. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline, which enhances LLMs' ability to follow constraints. DeCRIM works by decomposing the original instruction into a list of constraints and using a Critic model to decide when and where the LLM's response needs refinement. Our results show that DeCRIM improves Mistral's performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback. Moreover, we demonstrate that with strong feedback, open-source LLMs with DeCRIM can outperform GPT-4 on both benchmarks., Comment: To appear at EMNLP 2024

The process of creating training data to teach models is currently driven by humans, who manually analyze model weaknesses and plan how to create data that improves a student model. Recent approaches using LLMs as annotators reduce human effort, but still require humans to interpret feedback from evaluations and control the LLM to produce data the student needs. Automating this labor-intensive process by creating autonomous data generation agents - or teachers - is desirable, but requires environments that can simulate the feedback-driven, iterative, closed loop of data creation. To enable rapid and scalable testing for such agents and their modules, we introduce DataEnvGym, a testbed of teacher environments for data generation agents. DataEnvGym frames data generation as a sequential decision-making task, involving an agent consisting of a data generation policy (which generates a plan for creating training data) and a data generation engine (which transforms the plan into data), inside an environment that provides student feedback. The agent's goal is to improve student performance. Students are iteratively trained and evaluated on generated data, with their feedback (in the form of errors or weak skills) being reported to the agent after each iteration. DataEnvGym includes multiple teacher environment instantiations across 3 levels of structure in the state representation and action space. More structured environments are based on inferred skills and offer more interpretability and curriculum control. We support 3 diverse tasks (math, code, and VQA) and test multiple students and teachers. Example agents in our teaching environments can iteratively improve students across tasks and settings. Moreover, we show that environments teach different skill levels and test variants of key modules, pointing to future work in improving data generation agents, engines, and feedback mechanisms., Comment: Project Page: https://DataEnvGym.github.io