Search

Neural models for amortized probabilistic clustering yield samples of cluster labels given a set-structured input, while avoiding lengthy Markov chain runs and the need for explicit data likelihoods. Existing methods which label each data point sequentially, like the Neural Clustering Process, often lead to cluster assignments highly dependent on the data order. Alternatively, methods that sequentially create full clusters, do not provide assignment probabilities. In this paper, we introduce GFNCP, a novel framework for amortized clustering. GFNCP is formulated as a Generative Flow Network with a shared energy-based parametrization of policy and reward. We show that the flow matching conditions are equivalent to consistency of the clustering posterior under marginalization, which in turn implies order invariance. GFNCP also outperforms existing methods in clustering performance on both synthetic and real-world data., Comment: Accepted to AISTATS 2025 on January 21, 2025

The growing prevalence of Large Language Models (LLMs) is reshaping online text-based communication; a transformation that is extensively studied as AI-mediated communication. However, much of the existing research remains bound by traditional communication models, where messages are created and transmitted directly between humans despite LLMs being able to play a more active role in transforming messages. In this work, we propose the Intersubjective Model of AI-mediated Communication, an alternative communication model that leverages LLM-based adaptive agents to augment human-human communication. Unlike traditional communication models that focus on the accurate transmission of information, the Intersubjective Model allows for communication to be designed in an adaptive and customizable way to create alternative interactions by dynamically shaping messages in real time and facilitating shared understanding between the human participants. In this paper, we have developed a prototype text chat system based on the Intersubjective Model to describe the potential of this model, as well as the design space it affords.

The modern theory of polarization does not apply in its original form to systems with non-trivial band topology. Chern insulators are one such example since they are insulating in the bulk but exhibit metallic edge states, complicating the definition of polarization. Wannier functions formed a key ingredient of the original modern theory of polarization, but it has been considered that these cannot be applied to Chern insulators since they are no longer exponentially localized and the Wannier center is no longer gauge invariant. In this Letter, we provide an unambiguous definition of absolute polarization for a Chern insulator in terms of the Zak phase. We obtain our expression by studying the non-quantized fractional charge bound to lattice dislocations and it can be computed directly from bulk quantities and makes no assumption on the edge state filling. Our result is fully consistent with previous results on the quantized charge bound to dislocations in the presence of crystalline symmetry. At the same time, our result is more general since it also applies to Chern insulators which do not have crystalline symmetries other than translations., Comment: 5+5 pages, 2+3 figures

Augmented Reality (AR) is increasingly positioned as a tool for knowledge work, providing beneficial affordances such as a virtually limitless display space that integrates digital information with the user's physical surroundings. However, for AR to supplant traditional screen-based devices in knowledge work, it must support prolonged usage across diverse contexts. Until now, few studies have explored the effects, opportunities, and challenges of working in AR outside a controlled laboratory setting and for an extended duration. This gap in research limits our understanding of how users may adapt its affordances to their daily workflows and what barriers hinder its adoption. In this paper, we present findings from a longitudinal diary study examining how participants incorporated an AR laptop -- Sightful's Spacetop EA -- into their daily work routines. 14 participants used the device for 40-minute daily sessions over two weeks, collectively completing 103 hours of AR-based work. Through survey responses, workspace photographs, and post-study interviews, we analyzed usage patterns, workspace configurations, and evolving user perceptions. Our findings reveal key factors influencing participants' usage of AR, including task demands, environmental constraints, social dynamics, and ergonomic considerations. We highlight how participants leveraged and configured AR's virtual display space, along with emergent hybrid workflows that involved physical screens and tasks. Based on our results, we discuss both overlaps with current literature and new considerations and challenges for the future design of AR systems for pervasive and productive use.

When the game Lights Out is played according to an algorithm specifying the player's exact sequence of moves, it can be modeled using deterministic cellular automata. One such model reduces to the $\sigma$ automaton, which evolves according to the 2-dimensional analog of Rule 90. We consider how the cycle lengths of multi-dimensional $\sigma$ automata depend on their dimension. The main result of this work is that the cycle-lengths of 1-dimensional $\sigma$ automata and 2-dimensional $\sigma$ automata (of the same size) are equal, and we prove this by relating the eigenvalues and Jordan blocks of their respective transition matrices. We also find that cycle-lengths of higher-dimensional $\sigma$ automata are bounded (despite the number of lattice sites increasing with dimension) and eventually saturate the upper bound. On the way, we derive a general formula for the size of the largest Jordan block of the Kronecker sum of two matrices over $GF(2)$ using properties of Pascal's Triangle., Comment: 31 pages, 5 figures, 1 table

Recent progress has been made in detecting early stage dementia entirely through recordings of patient speech. Multimodal speech analysis methods were applied to the PROCESS challenge, which requires participants to use audio recordings of clinical interviews to predict patients as healthy control, mild cognitive impairment (MCI), or dementia and regress the patient's Mini-Mental State Exam (MMSE) scores. The approach implemented in this work combines acoustic features (eGeMAPS and Prosody) with embeddings from Whisper and RoBERTa models, achieving competitive results in both regression (RMSE: 2.7666) and classification (Macro-F1 score: 0.5774) tasks. Additionally, a novel two-tiered classification setup is utilized to better differentiate between MCI and dementia. Our approach achieved strong results on the test set, ranking seventh on regression and eleventh on classification out of thirty-seven teams, exceeding the baseline results., Comment: Submitted to IEEE ICAD 2025

Co-speech gesture generation is crucial for creating lifelike avatars and enhancing human-computer interactions by synchronizing gestures with speech. Despite recent advancements, existing methods struggle with accurately identifying the rhythmic or semantic triggers from audio for generating contextualized gesture patterns and achieving pixel-level realism. To address these challenges, we introduce Contextual Gesture, a framework that improves co-speech gesture video generation through three innovative components: (1) a chronological speech-gesture alignment that temporally connects two modalities, (2) a contextualized gesture tokenization that incorporate speech context into motion pattern representation through distillation, and (3) a structure-aware refinement module that employs edge connection to link gesture keypoints to improve video generation. Our extensive experiments demonstrate that Contextual Gesture not only produces realistic and speech-aligned gesture videos but also supports long-sequence generation and video gesture editing applications, shown in Fig.1 Project Page: https://andypinxinliu.github.io/Contextual-Gesture/.

A direct measurement of the ground-state-to-ground-state electron-capture decay $Q$ ($Q_{\rm EC}$) value of $^{97}$Tc has been conducted employing the high resolving power phase-imaging ion-cyclotron-resonance technique with the double Penning trap mass spectrometer JYFLTRAP. The resulting $Q_{\rm EC}$ value for $^{97}$Tc is 324.82(21) keV, exhibiting a precision approximately 19 times higher than the value adopted in the newest Atomic Mass Evaluation (AME2020) and differing by 1.2$\sigma$. Furthermore, by combining this refined $Q$ value with nuclear energy-level data for the decay-daughter $^{97}$Mo, a potential ultra-low Q-value transition, possibly of allowed type, $^{97}$Tc (9/2$^{+}$, ground state) $\rightarrow$ $^{97}$Mo$^{*}$ (320(1) keV), was evaluated for future long-term neutrino-mass determination experiments. The ground-state-to-excited-state electron-capture decay $Q$ value ($Q^{*}_{\rm EC}$) of this transition was determined to be 4.8(10) keV, confirming it to be energetically allowed with a confidence level of exceeding 4$\sigma$. The captures of electrons occupying the L and higher shells for this transition are energetically allowed, giving a value of 2.0(10) keV for the closest distance of $Q^{*}_{\rm EC}$ to the allowed binding energy of the L1 shell. To predict partial half-lives and energy-release distributions for this transition, the atomic self-consistent many-electron Dirac--Hartree--Fock--Slater method and the nuclear shell model have been employed. Dominant correction terms such as exchange and overlap corrections, as well as shake-up and shake-off effects, were included in the final results. Moreover, the normalized distribution of released energy in the electron-capture decay of $^{97}$Tc to excited states of $^{97}$Mo, is compared with that of $^{163}$Ho, which is being used for electron-neutrino-mass determination., Comment: 10 pages, 5 figures

The emergence of foundational models has greatly improved performance across various downstream tasks, with fine-tuning often yielding even better results. However, existing fine-tuning approaches typically require access to model weights and layers, leading to challenges such as managing multiple model copies or inference pipelines, inefficiencies in edge device optimization, and concerns over proprietary rights, privacy, and exposure to unsafe model variants. In this paper, we address these challenges by exploring "Gray-box" fine-tuning approaches, where the model's architecture and weights remain hidden, allowing only gradient propagation. We introduce a novel yet simple and effective framework that adapts to new tasks using two lightweight learnable modules at the model's input and output. Additionally, we present a less restrictive variant that offers more entry points into the model, balancing performance with model exposure. We evaluate our approaches across several backbones on benchmarks such as text-image alignment, text-video alignment, and sketch-image alignment. Results show that our Gray-box approaches are competitive with full-access fine-tuning methods, despite having limited access to the model.

We show that for any quadratic extension of number fields $K/F$, there exists an abelian variety $A/F$ of positive rank whose rank does not grow upon base change to $K$. This result implies that Hilbert's tenth problem over the ring of integers of any number field has a negative solution. That is, for the ring $\mathcal{O}_K$ of integers of any number field $K$, there does not exist an algorithm that answers the question of whether a polynomial equation in several variables over $\mathcal{O}_K$ has solutions in $\mathcal{O}_K$., Comment: 10 pages, comments welcome

We conduct the first systematic survey of a total of eleven optical coronal lines in the spectra of a large sample of low redshift (z < 0.8) Type 1 quasars observed by the Sloan Digital Sky Survey (SDSS). We find that strong coronal line emission is rare in SDSS even in Type 1 quasars; only 885 out of 19,508 (4.5%) galaxies show at least one coronal line, with higher ionization potential lines ($>100$eV) being even rarer. The [Ne V] $\lambda$3426 line, which constitutes the majority of detections, is strongly correlated with the bolometric luminosity. These findings suggest that the optical coronal lines are significantly suppressed in the majority of local AGNs, possibly as a result of the presence of dust in the emitting regions. We find that the incidence of ionized outflows is significantly higher in coronal line emitters compared with non-coronal line emitters, possibly suggesting that dust destruction in outflows enhances coronal line emission in AGNs. Many coronal lines show line profiles that are broader than those of narrow lines, and are blue-shifted relative the lower ionization potential lines, suggesting outflows in the highly ionized gas. Given the limited number of detections, we do not find any statistically significant trends of detection statistics, or line ratios with black hole mass, Eddington ratio, or AGN bolometric luminosity. The catalog is publicly available and can provide a useful database of the coronal line properties of low redshift quasars that can be compared to the growing number of high-z AGNs discovered by JWST., Comment: Accepted for publication in ApJS

3D printing has progressed significantly, allowing objects to be produced using a wide variety of materials. Recent advances have employed focused ultrasound in 3D printing, to allow printing inside acoustically transparent materials. Here we introduce a Selective Ultrasonic Melting (SUM) method for 3D printing of poly ({\epsilon}-caprolactone) (PCL) powder mixed with water. The printing was done by mechanically moving a focused ultrasound transducer. The microstructure and porosity of the prints were analyzed with micro-computed tomography ({\mu}CT). The open porosity of the printed samples was determined using the water intrusion method and by passing fluorescent microspheres through the structure. The cytocompatibility of the printed structures was confirmed by seeding NIH-3T3 fibroblast cells on the scaffolds, followed by analysis using live/dead fluorescent assay. and visualization using scanning electron microscopy (SEM). We demonstrated that SUM is a viable technique to print structures with active control of their porosity This method provides an alternative to methods such as fused deposition modelling (FDM) and material jetting.

Spectral Embedding (SE) is a popular method for dimensionality reduction, applicable across diverse domains. Nevertheless, its current implementations face three prominent drawbacks which curtail its broader applicability: generalizability (i.e., out-of-sample extension), scalability, and eigenvectors separation. In this paper, we introduce GrEASE: Generalizable and Efficient Approximate Spectral Embedding, a novel deep-learning approach designed to address these limitations. GrEASE incorporates an efficient post-processing step to achieve eigenvectors separation, while ensuring both generalizability and scalability, allowing for the computation of the Laplacian's eigenvectors on unseen data. This method expands the applicability of SE to a wider range of tasks and can enhance its performance in existing applications. We empirically demonstrate GrEASE's ability to consistently approximate and generalize SE, while ensuring scalability. Additionally, we show how GrEASE can be leveraged to enhance existing methods. Specifically, we focus on UMAP, a leading visualization technique, and introduce NUMAP, a generalizable version of UMAP powered by GrEASE. Our codes are publicly available.

In this work, we present new data on the $^{197}$Au photoneutron reactions in and above the giant dipole resonance region, obtained by using 8 to 39~MeV quasi-monochromatic $\gamma$-ray beams produced at the NewSUBARU facility in Japan and a high-and-flat efficiency neutron detection system. We report absolute cross sections and mean photoneutron energies for the $^{197}$Au$(\gamma,\,inX)$ reactions with $i$~=~1 to 4. The photoabsorption cross section was obtained as the sum of the $(\gamma,\,inX)$ reaction cross sections. The giant dipole resonance parameter values were obtained by fitting the experimental photoabsorption cross sections. The present photoabsorption cross sections are in good agreement with the Saclay results of Veyssiere~\emph{et al.}. Thus, our study does not support the recommendation of Berman~\emph{et al.} of lowering the Saclay photoabsorption cross sections by 8$\%$. We observed a non-statistical high-energy neutron emission in the $(\gamma,\,n)$ reaction in the low-energy region between $S_n$ and 10~MeV. The present results are compared with data from the literature and statistical model calculations performed with the TALYS and EMPIRE codes., Comment: arXiv admin note: text overlap with arXiv:2403.11547

Two seminal papers--Alon, Livni, Malliaris, Moran (STOC 2019) and Bun, Livni, and Moran (FOCS 2020)--established the equivalence between online learnability and globally stable PAC learnability in binary classification. However, Chase, Chornomaz, Moran, and Yehudayoff (STOC 2024) recently showed that this equivalence does not hold in the agnostic setting. Specifically, they proved that in the agnostic setting, only finite hypothesis classes are globally stable learnable. Therefore, agnostic global stability is too restrictive to capture interesting hypothesis classes. To address this limitation, Chase \emph{et al.} introduced two relaxations of agnostic global stability. In this paper, we characterize the classes that are learnable under their proposed relaxed conditions, resolving the two open problems raised in their work. First, we prove that in the setting where the stability parameter can depend on the excess error (the gap between the learner's error and the best achievable error by the hypothesis class), agnostic stability is fully characterized by the Littlestone dimension. Consequently, as in the realizable case, this form of learnability is equivalent to online learnability. As part of the proof of this theorem, we strengthen the celebrated result of Bun et al. by showing that classes with infinite Littlestone dimension are not stably PAC learnable, even if we allow the stability parameter to depend on the excess error. For the second relaxation proposed by Chase et al., we prove that only finite hypothesis classes are globally stable learnable even if we restrict the agnostic setting to distributions with small population loss.

We present analytical results for the distribution of first return (FR) times of non-backtracking random walks (NBWs) on undirected configuration model networks consisting of $N$ nodes with degree distribution $P(k)$. We focus on the case in which the network consists of a single connected component. Starting from a random initial node $i$ at time $t=0$, an NBW hops into a random neighbor of $i$ at time $t=1$ and at each subsequent step it continues to hop into a random neighbor of its current node, excluding the previous node. We calculate the tail distribution $P ( T_{\rm FR} > t )$ of first return times from a random node to itself. It is found that $P ( T_{\rm FR} > t )$ is given by a discrete Laplace transform of the degree distribution $P(k)$. This result exemplifies the relation between structural properties of a network, captured by the degree distribution, and properties of dynamical processes taking place on the network. Using the tail-sum formula, we calculate the mean first return time ${\mathbb E}[ T_{\rm FR} ]$. Surprisingly, ${\mathbb E}[ T_{\rm FR} ]$ coincides with the result obtained from the Kac's lemma that applies to classical random walks (RWs). We also calculate the variance ${\rm Var}(T_{\rm FR})$, which accounts for the variability of first return times between different NBW trajectories. We apply this formalism to random regular graphs, Erdos-R\'enyi networks and configuration model networks with exponential and power-law degree distributions and obtain closed-form expressions for $P ( T_{\rm FR} > t )$ as well as its mean and variance. These results provide useful insight on the advantages of NBWs over classical RWs in network exploration, sampling and search processes., Comment: 28 pages, 9 figures

We introduce the Byte Latent Transformer (BLT), a new byte-level LLM architecture that, for the first time, matches tokenization-based LLM performance at scale with significant improvements in inference efficiency and robustness. BLT encodes bytes into dynamically sized patches, which serve as the primary units of computation. Patches are segmented based on the entropy of the next byte, allocating more compute and model capacity where increased data complexity demands it. We present the first FLOP controlled scaling study of byte-level models up to 8B parameters and 4T training bytes. Our results demonstrate the feasibility of scaling models trained on raw bytes without a fixed vocabulary. Both training and inference efficiency improve due to dynamically selecting long patches when data is predictable, along with qualitative improvements on reasoning and long tail generalization. Overall, for fixed inference costs, BLT shows significantly better scaling than tokenization-based models, by simultaneously growing both patch and model size.

As superconducting kinetic inductance detectors (KIDs) continue to grow in popularity for sensitive sub-mm detection and other applications, there is a drive to advance toward lower loss devices. We present measurements of diagnostic thin film aluminum coplanar waveguide (CPW) resonators designed to inform ongoing KID development at NASA Goddard Space Flight Center. The resonators span $\rm f_0 = 3.5 - 4$\,GHz and include both quarter-wave and half-wave resonators with varying coupling capacitor designs. We present measurements of the device film properties and an analysis of the dominant mechanisms of loss in the resonators measured in a dark environment. We demonstrate quality factors of $\rm Q_i^{-1} \approx 3.64 - 8.57 \times10^{-8}$, and observe enhanced suppression of two-level system (TLS) loss in our devices at high internal microwave power levels before the onset of quasiparticle dissipation from microwave heating. We observe deviations from the standard TLS loss model at low powers and temperatures below 60 mK, and use a modified model to describe this behavior., Comment: 8 pages, 5 figures, submitted to Physical Review Applied Letters

Estimating uncertainty in Large Language Models (LLMs) is important for properly evaluating LLMs, and ensuring safety for users. However, prior approaches to uncertainty estimation focus on the final answer in generated text, ignoring intermediate steps that might dramatically impact the outcome. We hypothesize that there exist key forking tokens, such that re-sampling the system at those specific tokens, but not others, leads to very different outcomes. To test this empirically, we develop a novel approach to representing uncertainty dynamics across individual tokens of text generation, and applying statistical models to test our hypothesis. Our approach is highly flexible: it can be applied to any dataset and any LLM, without fine tuning or accessing model weights. We use our method to analyze LLM responses on 7 different tasks across 4 domains, spanning a wide range of typical use cases. We find many examples of forking tokens, including surprising ones such as punctuation marks, suggesting that LLMs are often just a single token away from saying something very different.

Deep neural networks exhibit empirical neural scaling laws, with error decreasing as a power law with increasing model or data size, across a wide variety of architectures, tasks, and datasets. This universality suggests that scaling laws may result from general properties of natural learning tasks. We develop a mathematical model intended to describe natural datasets using percolation theory. Two distinct criticality regimes emerge, each yielding optimal power-law neural scaling laws. These regimes, corresponding to power-law-distributed discrete subtasks and a dominant data manifold, can be associated with previously proposed theories of neural scaling, thereby grounding and unifying prior works. We test the theory by training regression models on toy datasets derived from percolation theory simulations. We suggest directions for quantitatively predicting language model scaling., Comment: 24 pages, 10 figures, Code: https://github.com/aribrill/scaling-laws-paper

The kinetic stability of collisionless, sloshing beam-ion (45{\deg} pitch angle) plasma is studied in a 3D simple magnetic mirror, mimicking the Wisconsin High-temperature superconductor Axisymmetric Mirror (WHAM) experiment. The collisional Fokker-Planck code CQL3D-m provides a slowing-down beam-ion distribution to initialize the kinetic-ion/fluid-electron code Hybrid-VPIC, which then simulates free plasma decay without external heating or fueling. Over 1-10 $\mu$s, drift-cyclotron loss-cone (DCLC) modes grow and saturate in amplitude. DCLC scatters ions to a marginally-stable distribution with gas-dynamic rather than classical-mirror confinement. Sloshing ions can trap cool (low-energy) ions in an electrostatic potential well to stabilize DCLC, but DCLC itself does not scatter sloshing beam-ions into said well. Instead, cool ions must come from external sources such as charge-exchange collisions with a low-density neutral population. Manually adding cool ~1 keV ions improves beam-ion confinement ~2-5x in Hybrid-VPIC simulations, which qualitatively corroborates measurements from real mirror devices with sloshing ions., Comment: Submitted; 35 pages, 14 figures

Inherent in the world of cryptocurrency systems and their security models is the notion that private keys, and thus assets, are controlled by individuals or individual entities. We present Liquefaction, a wallet platform that demonstrates the dangerous fragility of this foundational assumption by systemically breaking it. Liquefaction uses trusted execution environments (TEEs) to encumber private keys, i.e., attach rich, multi-user policies to their use. In this way, it enables the cryptocurrency credentials and assets of a single end-user address to be freely rented, shared, or pooled. It accomplishes these things privately, with no direct on-chain traces. Liquefaction demonstrates the sweeping consequences of TEE-based key encumbrance for the cryptocurrency landscape. Liquefaction can undermine the security and economic models of many applications and resources, such as locked tokens, DAO voting, airdrops, loyalty points, soulbound tokens, and quadratic voting. It can do so with no on-chain and minimal off-chain visibility. Conversely, we also discuss beneficial applications of Liquefaction, such as privacy-preserving, cost-efficient DAOs and a countermeasure to dusting attacks. Importantly, we describe an existing TEE-based tool that applications can use as a countermeasure to Liquefaction. Our work prompts a wholesale rethinking of existing models and enforcement of key and asset ownership in the cryptocurrency ecosystem., Comment: 19 pages, 10 figures

Active Learning (AL) is a user-interactive approach aimed at reducing annotation costs by selecting the most crucial examples to label. Although AL has been extensively studied for image classification tasks, the specific scenario of interactive image retrieval has received relatively little attention. This scenario presents unique characteristics, including an open-set and class-imbalanced binary classification, starting with very few labeled samples. We introduce a novel batch-mode Active Learning framework named GAL (Greedy Active Learning) that better copes with this application. It incorporates a new acquisition function for sample selection that measures the impact of each unlabeled sample on the classifier. We further embed this strategy in a greedy selection approach, better exploiting the samples within each batch. We evaluate our framework with both linear (SVM) and non-linear MLP/Gaussian Process classifiers. For the Gaussian Process case, we show a theoretical guarantee on the greedy approximation. Finally, we assess our performance for the interactive content-based image retrieval task on several benchmarks and demonstrate its superiority over existing approaches and common baselines. Code is available at https://github.com/barleah/GreedyAL., Comment: Accepted to Transactions on Machine Learning Research (TMLR)

The popularity of large language models (LLMs) continues to increase, and LLM-based assistants have become ubiquitous, assisting people of diverse backgrounds in many aspects of life. Significant resources have been invested in the safety of LLMs and their alignment with social norms. However, research examining their behavior from the information security awareness (ISA) perspective is lacking. Chatbots and LLM-based assistants may put unwitting users in harm's way by facilitating unsafe behavior. We observe that the ISA inherent in some of today's most popular LLMs varies significantly, with most models requiring user prompts with a clear security context to utilize their security knowledge and provide safe responses to users. Based on this observation, we created a comprehensive set of 30 scenarios to assess the ISA of LLMs. These scenarios benchmark the evaluated models with respect to all focus areas defined in a mobile ISA taxonomy. Among our findings is that ISA is mildly affected by changing the model's temperature, whereas adjusting the system prompt can substantially impact it. This underscores the necessity of setting the right system prompt to mitigate ISA weaknesses. Our findings also highlight the importance of ISA assessment for the development of future LLM-based assistants.

Runge-Kutta methods are affine equivariant: applying a method before or after an affine change of variables yields the same numerical trajectory. However, for some applications, one would like to perform numerical integration after a quadratic change of variables. For example, in Lie-Poisson reduction, a quadratic transformation reduces the number of variables in a Hamiltonian system, yielding a more efficient representation of the dynamics. Unfortunately, directly applying a symplectic Runge-Kutta method to the reduced system generally does not preserve its Hamiltonian structure, so many proposed techniques require computing numerical trajectories of the original, unreduced system. In this paper, we study when a Runge-Kutta method in the original variables descends to a numerical integrator expressible entirely in terms of the quadratically transformed variables. In particular, we show that symplectic diagonally implicit Runge-Kutta (SyDIRK) methods, applied to a quadratic projectable vector field, are precisely the Runge-Kutta methods that descend to a method (generally not of Runge-Kutta type) in the projected variables. We illustrate our results with several examples in both conservative and non-conservative dynamics., Comment: 21 pages

We show that the $j$-th Dirichlet eigenvalue of the sub-Laplacian on an open set of a Carnot group is greater than the $(j+1)$-st Neumann eigenvalue. This extends earlier results in the Euclidean and Heisenberg case and has a remarkably simple proof., Comment: 10 pages; dedicated to Nikolai Nadirashvili, on the occasion of his 70th birthday

Indonesia is one of the most diverse countries linguistically. However, despite this linguistic diversity, Indonesian languages remain underrepresented in Natural Language Processing (NLP) research and technologies. In the past two years, several efforts have been conducted to construct NLP resources for Indonesian languages. However, most of these efforts have been focused on creating manual resources thus difficult to scale to more languages. Although many Indonesian languages do not have a web presence, locally there are resources that document these languages well in printed forms such as books, magazines, and newspapers. Digitizing these existing resources will enable scaling of Indonesian language resource construction to many more languages. In this paper, we propose an alternative method of creating datasets by digitizing documents, which have not previously been used to build digital language resources in Indonesia. DriveThru is a platform for extracting document content utilizing Optical Character Recognition (OCR) techniques in its system to provide language resource building with less manual effort and cost. This paper also studies the utility of current state-of-the-art LLM for post-OCR correction to show the capability of increasing the character accuracy rate (CAR) and word accuracy rate (WAR) compared to off-the-shelf OCR., Comment: 12 pages, 3 figures, 6 tables

We propose protected pipelines or props for short, a new approach for authenticated, privacy-preserving access to deep-web data for machine learning (ML). By permitting secure use of vast sources of deep-web data, props address the systemic bottleneck of limited high-quality training data in ML development. Props also enable privacy-preserving and trustworthy forms of inference, allowing for safe use of sensitive data in ML applications. Props are practically realizable today by leveraging privacy-preserving oracle systems initially developed for blockchain applications.

Scheduling languages express to a compiler a sequence of optimizations to apply. Compilers that support a scheduling language interface allow exploration of compiler optimizations, i.e., exploratory compilers. While scheduling languages have become a common feature of tools for expert users, the proliferation of these languages without unifying common features may be confusing to users. Moreover, we recognize a need to organize the compiler developer community around common exploratory compiler infrastructure, and future advances to address, for example, data layout and data movement. To support a broader set of users may require raising the level of abstraction. This paper provides a taxonomy of scheduling languages, first discussing their origins in iterative compilation and autotuning, noting the common features and how they are used in existing frameworks, and then calling for changes to increase their utility and portability.

We provide algorithms to decide whether a finitely generated subgroup of $\mathrm{SL}_2(\mathbb{R})$ is discrete, solve the constructive membership problem for finitely generated discrete subgroups of $\mathrm{SL}_2(\mathbb{R})$, and compute a fundamental domain for a finitely generated Fuchsian group. These algorithms have been implemented in Magma for groups defined over real algebraic number fields., Comment: 27 pages, 14 figures

We study a setting where an intelligent model (e.g., a pre-trained neural network) predicts the real-time value of a target signal using data samples transmitted from a remote source according to a scheduling policy. The scheduler decides on i) the age of the samples to be sent, ii) when to send them, and iii) the length of each packet (i.e., the number of samples contained in each packet). The dependence of inference quality on the Age of Information (AoI) for a given packet length is modeled by a general relationship. Previous work assumed i.i.d. transmission delays with immediate feedback or were restricted to the case where inference performance degrades as the input data ages. Our formulation, in addition to capturing non-monotone age dependence, also covers Markovian delay on both forward and feedback links. We model this as an infinite-horizon average-cost Semi-Markov Decision Process. We obtain a closed-form solution that decides on (i) and (ii) for any constant packet length. The solution for when to send is an index-based threshold policy, where the index function is expressed in terms of the delay state and AoI at the receiver. The age of the packet selected is a function of the delay state. We separately optimize the value of the constant length. We also develop an index-based threshold policy for the variable length case, which allows a complexity reduction. In simulation results, we observe that our goal-oriented scheduler drops inference error down to one sixth with respect to age-based scheduling of unit-length packets., Comment: 13 pages, 9 figures

Emergency department (ED) crowding is a global public health issue that has been repeatedly associated with increased mortality. Predicting future service demand would enable preventative measures aiming to eliminate crowding along with it's detrimental effects. Recent findings in our ED indicate that occupancy ratios exceeding 90% are associated with increased 10-day mortality. In this paper, we aim to predict these crisis periods using retrospective data from a large Nordic ED with a LightGBM model. We provide predictions for the whole ED and individually for it's different operational sections. We demonstrate that afternoon crowding can be predicted at 11 a.m. with an AUC of 0.82 (95% CI 0.78-0.86) and at 8 a.m. with an AUC up to 0.79 (95% CI 0.75-0.83). Consequently we show that forecasting mortality-associated crowding using anonymous administrative data is feasible.

We present the pc-scale radio spectra of a representative sample of 13 Palomar-Green radio-quiet quasars, based on our new Very Long Baseline Array (VLBA) observations at 8.4 and 23.6 GHz and our earlier VLBA studies at 1.5 and 5.0 GHz. The radio core emission generally exhibits a flat spectrum at 1.5-5.0 GHz, which indicates a compact optically thick synchrotron source on a scale smaller than the broad-line region (BLR) radius R_BLR ~ 0.01-0.1 pc. The 8.4-23.6 GHz spectral slope remains flat in four objects indicating the inner radius of the radio source R_in < 0.1 R_BLR, and becomes steep in four other objects indicating R_in ~ 0.5 R_BLR. The flat 8.4-23.6 GHz slope sources may be associated with a continuous ejection starting at the accretion disk corona. The steep 8.4-23.6 GHz slope sources may be produced by an interaction of an AGN-driven wind with the BLR gas or a low-power jet extending to the BLR scale. Seven of these eight objects, which have a flat or steep 8.4-23.6 GHz slope, reside at the Eddington ratios L/L_Edd < 0.3, and four of the remaining five objects, where the 8.4-23.6 GHz fluxes are too faint to significantly constrain the slope, reside at L/L_Edd > 0.3. The 8.4-23.6 GHz radio emission in the high L/L_Edd objects may be weak due to more extended emission from a radiation pressure driven wind. Future sub-millimeter observations can further constrain the inward radial extent of the radio emission down to the coronal scale., Comment: Accepted for publication in ApJ

Resolving knowledge conflicts is a crucial challenge in Question Answering (QA) tasks, as the internet contains numerous conflicting facts and opinions. While some research has made progress in tackling ambiguous settings where multiple valid answers exist, these approaches often neglect to provide source citations, leaving users to evaluate the factuality of each answer. On the other hand, existing work on citation generation has focused on unambiguous settings with single answers, failing to address the complexity of real-world scenarios. Despite the importance of both aspects, no prior research has combined them, leaving a significant gap in the development of QA systems. In this work, we bridge this gap by proposing the novel task of QA with source citation in ambiguous settings, where multiple valid answers exist. To facilitate research in this area, we create a comprehensive framework consisting of: (1) five novel datasets, obtained by augmenting three existing reading comprehension datasets with citation meta-data across various ambiguous settings, such as distractors and paraphrasing; (2) the first ambiguous multi-hop QA dataset featuring real-world, naturally occurring contexts; (3) two new metrics to evaluate models' performances; and (4) several strong baselines using rule-based, prompting, and finetuning approaches over five large language models. We hope that this new task, datasets, metrics, and baselines will inspire the community to push the boundaries of QA research and develop more trustworthy and interpretable systems., Comment: Accepted to EMNLP 2024

A major challenge in autonomous vehicle research is modeling agent behaviors, which has critical applications including constructing realistic and reliable simulations for off-board evaluation and forecasting traffic agents motion for onboard planning. While supervised learning has shown success in modeling agents across various domains, these models can suffer from distribution shift when deployed at test-time. In this work, we improve the reliability of agent behaviors by closed-loop fine-tuning of behavior models with reinforcement learning. Our method demonstrates improved overall performance, as well as improved targeted metrics such as collision rate, on the Waymo Open Sim Agents challenge. Additionally, we present a novel policy evaluation benchmark to directly assess the ability of simulated agents to measure the quality of autonomous vehicle planners and demonstrate the effectiveness of our approach on this new benchmark.

The BICEP3 and BICEP Array polarimeters are small-aperture refracting telescopes located at the South Pole designed to measure primordial gravitational wave signatures in the Cosmic Microwave Background (CMB) polarization, predicted by inflation. Constraining the inflationary signal requires not only excellent sensitivity, but also careful control of instrumental systematics. Both instruments use antenna-coupled orthogonally polarized detector pairs, and the polarized sky signal is reconstructed by taking the difference in each detector pair. As a result, the differential response between detectors within a pair becomes an important systematic effect we must control. Additionally, mapping the intensity and polarization response in regions away from the main beam can inform how sidelobe levels affect CMB measurements. Extensive calibration measurements are taken in situ every austral summer for control of instrumental systematics and instrument characterisation. In this work, we detail the set of beam calibration measurements that we conduct on the BICEP receivers, from deep measurements of main beam response to polarized beam response and sidelobe mapping. We discuss the impact of these measurements for instrumental systematics studies and design choices for future CMB receivers., Comment: 13 pages, 7 figures, 1 table, Proceedings paper SPIE 2024

Class-agnostic image segmentation is a crucial component in automating image editing workflows, especially in contexts where object selection traditionally involves interactive tools. Existing methods in the literature often adhere to top-down formulations, following the paradigm of class-based approaches, where object detection precedes per-object segmentation. In this work, we present a novel bottom-up formulation for addressing the class-agnostic segmentation problem. We supervise our network directly on the projective sphere of its feature space, employing losses inspired by metric learning literature as well as losses defined in a novel segmentation-space representation. The segmentation results are obtained through a straightforward mean-shift clustering of the estimated features. Our bottom-up formulation exhibits exceptional generalization capability, even when trained on datasets designed for class-based segmentation. We further showcase the effectiveness of our generic approach by addressing the challenging task of cell and nucleus segmentation. We believe that our bottom-up formulation will offer valuable insights into diverse segmentation challenges in the literature.

In order to solve problems of practical importance, quantum computers will likely need to incorporate quantum error correction, where a logical qubit is redundantly encoded in many noisy physical qubits. The large physical-qubit overhead typically associated with error correction motivates the search for more hardware-efficient approaches. Here, using a microfabricated superconducting quantum circuit, we realize a logical qubit memory formed from the concatenation of encoded bosonic cat qubits with an outer repetition code of distance $d=5$. The bosonic cat qubits are passively protected against bit flips using a stabilizing circuit. Cat-qubit phase-flip errors are corrected by the repetition code which uses ancilla transmons for syndrome measurement. We realize a noise-biased CX gate which ensures bit-flip error suppression is maintained during error correction. We study the performance and scaling of the logical qubit memory, finding that the phase-flip correcting repetition code operates below threshold, with logical phase-flip error decreasing with code distance from $d=3$ to $d=5$. Concurrently, the logical bit-flip error is suppressed with increasing cat-qubit mean photon number. The minimum measured logical error per cycle is on average $1.75(2)\%$ for the distance-3 code sections, and $1.65(3)\%$ for the longer distance-5 code, demonstrating the effectiveness of bit-flip error suppression throughout the error correction cycle. These results, where the intrinsic error suppression of the bosonic encodings allows us to use a hardware-efficient outer error correcting code, indicate that concatenated bosonic codes are a compelling paradigm for reaching fault-tolerant quantum computation., Comment: Comments on the manuscript welcome!

A pandemic is the spread of a disease across large regions, and can have devastating costs to the society in terms of health, economic and social. As such, the study of effective pandemic mitigation strategies can yield significant positive impact on the society. A pandemic can be mathematically described using a compartmental model, such as the Susceptible Infected Removed (SIR) model. In this paper, we extend the solution equations of the SIR model to a state transition model with lockdowns. We formalize a metric hybrid planning problem based on this state transition model, and solve it using a metric hybrid planner. We improve the runtime effectiveness of the metric hybrid planner with the addition of valid inequalities, and demonstrate the success of our approach both theoretically and experimentally under various challenging settings.

The current event cameras are bio-inspired sensors that respond to brightness changes in the scene asynchronously and independently for every pixel, and transmit these changes as ternary event streams. Event cameras have several benefits over conventional digital cameras, such as significantly higher temporal resolution and pixel bandwidth resulting in reduced motion blur, and very high dynamic range. However, they also introduce challenges such as the difficulty of applying existing computer vision algorithms to the output event streams, and the flood of uninformative events in the presence of oscillating light sources. Here we propose a new type of event, the gradient event, which benefits from the same properties as a conventional brightness event, but which is by design much less sensitive to oscillating light sources, and which enables considerably better grayscale frame reconstruction. We show that the gradient event -based video reconstruction outperforms existing state-of-the-art brightness event -based methods by a significant margin, when evaluated on publicly available event-to-video datasets. Our results show how gradient information can be used to significantly improve the acquisition of visual information by an event camera., Comment: 8 pages, 6 figures. This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement no 101016734

Recent advancements in monocular neural depth estimation, particularly those achieved by the UniDepth network, have prompted the investigation of integrating UniDepth within a Gaussian splatting framework for monocular SLAM.This study presents UDGS-SLAM, a novel approach that eliminates the necessity of RGB-D sensors for depth estimation within Gaussian splatting framework. UDGS-SLAM employs statistical filtering to ensure local consistency of the estimated depth and jointly optimizes camera trajectory and Gaussian scene representation parameters. The proposed method achieves high-fidelity rendered images and low ATERMSE of the camera trajectory. The performance of UDGS-SLAM is rigorously evaluated using the TUM RGB-D dataset and benchmarked against several baseline methods, demonstrating superior performance across various scenarios. Additionally, an ablation study is conducted to validate design choices and investigate the impact of different network backbone encoders on system performance.

A natural strategy for continual learning is to weigh a Bayesian ensemble of fixed functions. This suggests that if a (single) neural network could be interpreted as an ensemble, one could design effective algorithms that learn without forgetting. To realize this possibility, we observe that a neural network classifier with N parameters can be interpreted as a weighted ensemble of N classifiers, and that in the lazy regime limit these classifiers are fixed throughout learning. We call these classifiers the neural tangent experts and show they output valid probability distributions over the labels. We then derive the likelihood and posterior probability of each expert given past data. Surprisingly, the posterior updates for these experts are equivalent to a scaled and projected form of stochastic gradient descent (SGD) over the network weights. Away from the lazy regime, networks can be seen as ensembles of adaptive experts which improve over time. These results offer a new interpretation of neural networks as Bayesian ensembles of experts, providing a principled framework for understanding and mitigating catastrophic forgetting in continual learning settings., Comment: Presented as a spotlight paper at NeurIPS, 2024

The performance and near-wake characteristics of a turbine in a confined flow depend on the blockage ratio, defined as the ratio of the turbine projected area to the channel cross-sectional area. While blockage is understood to increase the power coefficient for turbine "fences" spanning a channel, most investigations at the upper range of practically-achievable blockage ratios have been theoretical or numerical in nature. Furthermore, while linear momentum actuator disk theory is frequently used to model turbines in confined flows, as confinement increases, the ability of this idealized model to describe performance and flow fields has not been established. In this work, the performance and near-wake flow field of a pair of cross-flow turbines are experimentally evaluated at blockage ratios from 30% to 55%. The fluid velocity measured in the bypass region is found to be well-predicted by the open-channel linear momentum model developed by Houlsby et al. (2008), while the wake velocity is not. Additionally, self-similar power and thrust coefficients are identified across this range of blockage ratios when array performance is scaled by the modeled bypass velocity following Whelan et al.'s (2009) adaptation of the bluff-body theory of Maskell (1963). This result demonstrates that, despite multiple non-idealities, relatively simple models can quantitatively describe highly confined turbines. From this, an analytical method for predicting array performance as a function of blockage is presented. Overall, this work illustrates turbine performance at relatively high confinement and demonstrates the suitability of analytical models to predict and interpret their hydrodynamics., Comment: SUBMITTED to Journal of Fluid Mechanics. Data supporting this manuscript will soon be available at https://doi.org/10.5061/dryad.dfn2z35b5

The emergence of cognition requires a framework that bridges evolutionary principles with neurocomputational mechanisms. This paper introduces the novel "thoughtseed" framework, proposing that cognition arises from the dynamic interaction of self-organizing units of embodied knowledge called "thoughtseeds" within the Global Workspace of consciousness. Leveraging foundational concepts from evolutionary theory, neuronal packets, and free energy principle, we propose a hierarchical model of cognitive states, comprising Neuronal Packet Domains (NPDs), Knowledge Domains (KDs), the thoughtseed network, and meta-cognition. This hierarchical interplay, mediated by nested Markov blankets and reciprocal message passing, facilitates the emergence of thoughtseeds as coherent patterns of activity that guide perception, action, and learning. Thoughtseeds, posited as fundamental units of thought, compete for dominance within the Global Workspace, with the dominant thoughtseed shaping conscious experience and guiding behavior. We present a mathematical framework grounded in active inference and dynamical systems theory to model thoughtseed dynamics and their contribution to the unitary nature of consciousness. The thoughtseed framework offers a promising step towards a novel, biologically-grounded model for understanding the organizing principles and emergence of embodied cognition, offering a unified account of cognitive phenomena, with potential applications in understanding consciousness, attention, and decision-making., Comment: 2nd version. Reduced in size from 88 pages to 48 pages, to focus on the core framework. Items removed: Detailed discussions on Evolutionary Biology, contemplative traditions and computational theories of Mind. Replaced Inner Screen Model with Global Workspace Theory. Mathematical equations have been streamlined and is better organized. preprint arXiv:2408.15982

Does a broad-line region (BLR) wind in radio-quiet (RQ) active galactic nuclei (AGN) extend to pc scales and produce radio emission? We explore the correlations between a pc-scale radio wind and the BLR wind in a sample of 19 RQ Palomar-Green (PG) quasars. The radio wind is defined based on the spectral slope and the compactness of the emission at 1.5-5 GHz, and the BLR wind is defined by the excess blue wing in the C IV emission line profile. The five objects with both radio and BLR wind indicators are found at high Eddington ratios L/L_Edd (> 0.66), and eight of the nine objects with neither radio nor BLR winds reside at low L/L_Edd (< 0.28). This suggests that the BLR wind and the radio wind in RQ AGN are related to a radiation pressure driven wind. Evidence for free-free absorption by AGN photoionized gas, which flattens the spectral slope, is found in two objects. Radio outflows in three low L/L_Edd (0.05-0.12) objects are likely from a low-power jet, as suggested by additional evidence. The presence of a mild equatorial BLR wind in four intermediate L/L_Edd (0.2-0.4) objects can be tested with future spectropolarimetry., Comment: Accepted for publication in ApJ

Forbidden collisionally excited optical atomic transitions from high ionization potential (IP$\geq$54.8\,eV) ions, such as Ca$^{\mathrm{4+}}$, Ne$^{\mathrm{4+}}$, Fe$^{\mathrm{6+}}$, Fe$^{\mathrm{10+}}$, Fe$^{\mathrm{13+}}$, Ar$^{\mathrm{9+}}$, and S$^{\mathrm{11+}}$, are known as optical coronal lines (CLs). The spectral energy distribution (SED) of active galactic nuclei (AGN) typically extends to hundreds of electron volts and above, which should be able to produce such highly ionized gas. However, optical CLs are often not detected in AGN. Here we use photoionization calculations with the \textsc{Cloudy} spectral synthesis code to determine possible reasons for the rarity of these optical CLs. We calculate CL luminosities and equivalent widths from radiation pressure confined photoionized gas slabs exposed to an AGN continuum. We consider the role of dust, metallicity, and ionizing SED in the formation of optical CLs. We find that (1) dust reduces the strength of most CLs by $\sim$three orders of magnitude, primarily as a result of depletion of metals onto the dust grains. (2) In contrast to the CLs, the more widely observed lower IP optical lines such as [O\, III] 5007\,\AA, are less affected by depletion and some are actually enhanced in dusty gas. (3) In dustless gas many optical CLs become detectable, and are particularly strong for a hard ionizing SED. This implies that prominent CL emission likely originate in dustless gas. Our calculations also suggest optical CL emission is enhanced in galaxies with low mass black holes characterized by a harder radiation field and a low dust to metal ratio. The fact that optical CLs are not widely observed in the early universe with JWST may point to rapid dust formation at high redshift., Comment: 21 pages, 6 figures, accepted for publication in The Astrophysical Journal

The research was conducted at the College of Agricultural Sciences Engineering/University of Sulaimani/ Kurdistan Region-Iraqi to investigate effects of different concentrations of IBA (0, 3000, 4000 and 5000 ppm) and soaking in water for 24 hours on propagation black mulberry (Morus nigra L.) by hardwood cuttings. In this research the parameters of rooting percentage, root number, root length, sprout bud number, shoot length and shoot diameter were measured. Effect of individual factors showed that the highest rooting percentage (15%) was achieved in cuttings soaked in water for 24 hours, as well as improving other traits. Also, the best (23.33%) rooting was found in cuttings dipped in 4000 ppm IBA. Interaction effects of the two factors showed that cuttings treated with 4000 ppm IBA and soaked in water for 24 hours gave the highest (40%) rooting, and the highest other root and shoot traits were achieved in the same interaction as well.

Language Model (LM) agents for cybersecurity that are capable of autonomously identifying vulnerabilities and executing exploits have potential to cause real-world impact. Policymakers, model providers, and researchers in the AI and cybersecurity communities are interested in quantifying the capabilities of such agents to help mitigate cyberrisk and investigate opportunities for penetration testing. Toward that end, we introduce Cybench, a framework for specifying cybersecurity tasks and evaluating agents on those tasks. We include 40 professional-level Capture the Flag (CTF) tasks from 4 distinct CTF competitions, chosen to be recent, meaningful, and spanning a wide range of difficulties. Each task includes its own description, starter files, and is initialized in an environment where an agent can execute commands and observe outputs. Since many tasks are beyond the capabilities of existing LM agents, we introduce subtasks for each task, which break down a task into intermediary steps for a more detailed evaluation. To evaluate agent capabilities, we construct a cybersecurity agent and evaluate 8 models: GPT-4o, OpenAI o1-preview, Claude 3 Opus, Claude 3.5 Sonnet, Mixtral 8x22b Instruct, Gemini 1.5 Pro, Llama 3 70B Chat, and Llama 3.1 405B Instruct. For the top performing models (GPT-4o and Claude 3.5 Sonnet), we further investigate performance across 4 agent scaffolds (structed bash, action-only, pseudoterminal, and web search). Without subtask guidance, agents leveraging Claude 3.5 Sonnet, GPT-4o, OpenAI o1-preview, and Claude 3 Opus successfully solved complete tasks that took human teams up to 11 minutes to solve. In comparison, the most difficult task took human teams 24 hours and 54 minutes to solve. All code and data are publicly available at https://cybench.github.io., Comment: 151 pages, 9 figures

I present the first complete theory of OCD. OCD occurs when excessive CRH is released in the prefrontal cortex, activating cAMP. cAMP is a major inducer of HCN channels, which promote repeated neural firing. The combination of CRH, which is strongly associated with stress, and repeated firing that cannot be controlled, explains all of the features of OCD, including obsessions and compulsions of all kinds.

I present a theory of schizophrenia (SZ) that mechanistically explains its etiology, symptoms, pathophysiology, and treatment. SZ involves the chronic release of membrane polyunsaturated fatty acids (PUFAs) and their utilization for the synthesis of stress-induced plasticity agents such as endocannabinoids (ECBs). The causal event in SZ is prolonged stress during a sensitive period, which can induce prolonged and heritable changes. The physiological effect of the released PUFAs and their products is to disconnect neurons from their inputs and promote intrinsic excitability. I show that these effects can explain the positive, negative, cognitive, and mood symptoms of SZ, as well as the mechanisms of many known triggers of psychosis. The theory is supported by overwhelming evidence addressing lipids, immunity, ECBs, neuromodulators, hormones, neurotransmitters, and cortical parameters in SZ. It explains why antipsychotic drugs are effective against positive symptoms, and why they do not affect the other symptoms. Finally, I present promising treatment directions implied by the theory, including some that are immediately available.