Search

Large Language Models (LLMs) have shown remarkable capabilities across various natural language processing tasks but often struggle to excel uniformly in diverse or complex domains. We propose a novel ensemble method - Diverse Fingerprint Ensemble (DFPE), which leverages the complementary strengths of multiple LLMs to achieve more robust performance. Our approach involves: (1) clustering models based on response "fingerprints" patterns, (2) applying a quantile-based filtering mechanism to remove underperforming models at a per-subject level, and (3) assigning adaptive weights to remaining models based on their subject-wise validation accuracy. In experiments on the Massive Multitask Language Understanding (MMLU) benchmark, DFPE outperforms the best single model by 3% overall accuracy and 5% in discipline-level accuracy. This method increases the robustness and generalization of LLMs and underscores how model selection, diversity preservation, and performance-driven weighting can effectively address challenging, multi-faceted language understanding tasks.

Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai., Comment: 26 pages, 6 figures

Algorithmic decision-making has become deeply ingrained in many domains, yet biases in machine learning models can still produce discriminatory outcomes, often harming unprivileged groups. Achieving fair classification is inherently challenging, requiring a careful balance between predictive performance and ethical considerations. We present FairTTTS, a novel post-processing bias mitigation method inspired by the Tree Test Time Simulation (TTTS) method. Originally developed to enhance accuracy and robustness against adversarial inputs through probabilistic decision-path adjustments, TTTS serves as the foundation for FairTTTS. By building on this accuracy-enhancing technique, FairTTTS mitigates bias and improves predictive performance. FairTTTS uses a distance-based heuristic to adjust decisions at protected attribute nodes, ensuring fairness for unprivileged samples. This fairness-oriented adjustment occurs as a post-processing step, allowing FairTTTS to be applied to pre-trained models, diverse datasets, and various fairness metrics without retraining. Extensive evaluation on seven benchmark datasets shows that FairTTTS outperforms traditional methods in fairness improvement, achieving a 20.96% average increase over the baseline compared to 18.78% for related work, and further enhances accuracy by 0.55%. In contrast, competing methods typically reduce accuracy by 0.42%. These results confirm that FairTTTS effectively promotes more equitable decision-making while simultaneously improving predictive performance.

As machine learning (ML) systems increasingly impact critical sectors such as hiring, financial risk assessments, and criminal justice, the imperative to ensure fairness has intensified due to potential negative implications. While much ML fairness research has focused on enhancing training data and processes, addressing the outputs of already deployed systems has received less attention. This paper introduces 'BiasGuard', a novel approach designed to act as a fairness guardrail in production ML systems. BiasGuard leverages Test-Time Augmentation (TTA) powered by Conditional Generative Adversarial Network (CTGAN), a cutting-edge generative AI model, to synthesize data samples conditioned on inverted protected attribute values, thereby promoting equitable outcomes across diverse groups. This method aims to provide equal opportunities for both privileged and unprivileged groups while significantly enhancing the fairness metrics of deployed systems without the need for retraining. Our comprehensive experimental analysis across diverse datasets reveals that BiasGuard enhances fairness by 31% while only reducing accuracy by 0.09% compared to non-mitigated benchmarks. Additionally, BiasGuard outperforms existing post-processing methods in improving fairness, positioning it as an effective tool to safeguard against biases when retraining the model is impractical.

We present the results of the second commissioning phase of the short-focal-length area of the Apollon laser facility (located in Saclay, France), which was performed with the main laser beam (F1), scaled to a peak power of 2 PetaWatt. Under the conditions that were tested, this beam delivered on-target pulses of maximum energy up to 45 J and 22 fs duration. Several diagnostics were fielded to assess the performance of the facility. The on-target focal spot and its spatial stability, as well as the secondary sources produced when irradiating solid targets, have all been characterized, with the goal of helping users design future experiments. The laser-target interaction was characterized, as well as emissions of energetic ions, X-ray and neutrons recorded, all showing good laser-to-target coupling efficiency. Moreover, we demonstrated the simultaneous fielding of F1 with the auxiliary 0.5 PW F2 beam of Apollon, enabling dual beam operation. The present commissioning will be followed in 2025 by a further commissioning stage of F1 at the 8 PW level, en route to the final 10 PW goal.

The fall 2023 and spring 2024 undergraduate enrollment increases, marking the first growth since the COVID-19 pandemic, show signs of a post-pandemic turnaround for higher education. However, a significant share of current undergraduates will eventually disengage from college before earning a degree or other credential. They will join tens of millions of other adult Americans who are Some College, No Credential (SCNC). The SCNC population has been consistently rising over time. Re-engaging those who stop out remains a persistent challenge and a priority for the forty states that have set ambitious postsecondary attainment goals. This report aims to provide timely insights into the SCNC population, offering state leaders and policymakers accurate data on its current status, along with tracking progress and outcome measures for SCNC students. The first section of this report describes who makes up the SCNC population and how it has changed since the last report. In this section, the authors pay particular attention to Recent Stopouts, who joined the SCNC population after being stopped out between January 2021 and July 2022. In the second section, the authors report on SCNC re-enrollment in the 2022-23 academic year as well as first-year credential earning for re-enrollees. The authors also provide new updates on continued enrollment and second-year credential earning for SCNC re-enrollees in the 2021-22 academic year, whom were first reported on last year.

A distinct power-law break is apparent m_AB approximately 21 in the deep Near-Infrared PEARLS-JWST galaxy counts. The break becomes more pronounced at longer wavelengths, with the counts slope flattening smoothly with apparent magnitude in the shortest band used at 0.9 microns, trending towards an increasingly broken slope by the longest wavelength passband of JWST NIRCam, 4.4 microns. This behaviour is remarkably well predicted by the GALFORM semi-analytical model of galaxy formation. We use the model to diagnose the origin of this behaviour. We find that the features that are responsible for the break are: 1) the inherent break in the luminosity function; 2) the change in the volume element with redshift and 3) the redshift-dependent nature of the k-correction. We study the contribution to these effects by early and late-type galaxies, using as a proxy for morphology the bulge-to-total stellar mass ratio. We find that the way in which ellipticals populate the bright end of the luminosity function while spirals dominate the faint end is preserved in the galaxy number counts, with a characteristic stellar mass at the break of approximately 10^10 M_sun. We also find that the shape of the number counts is mainly driven by galaxies with relatively low redshift (z < 2) for the PEARLS observational limit of m_AB < 28. We give a comprehensive description of why the galaxy number counts in the near-infrared PEARLS-JWST observation look the way they do and which population of galaxies is dominant at each apparent magnitude.

Efforts to connect LiDAR data with text, such as LidarCLIP, have primarily focused on embedding 3D point clouds into CLIP text-image space. However, these approaches rely on 3D point clouds, which present challenges in encoding efficiency and neural network processing. With the advent of advanced LiDAR sensors like Ouster OS1, which, in addition to 3D point clouds, produce fixed resolution depth, signal, and ambient panoramic 2D images, new opportunities emerge for LiDAR based tasks. In this work, we propose an alternative approach to connect LiDAR data with text by leveraging 2D imagery generated by the OS1 sensor instead of 3D point clouds. Using the Florence 2 large model in a zero-shot setting, we perform image captioning and object detection. Our experiments demonstrate that Florence 2 generates more informative captions and achieves superior performance in object detection tasks compared to existing methods like CLIP. By combining advanced LiDAR sensor data with a large pre-trained model, our approach provides a robust and accurate solution for challenging detection scenarios, including real-time applications requiring high accuracy and robustness.

Recent advances in generative image restoration (IR) have demonstrated impressive results. However, these methods are hindered by their substantial size and computational demands, rendering them unsuitable for deployment on edge devices. This work introduces ELIR, an Efficient Latent Image Restoration method. ELIR operates in latent space by first predicting the latent representation of the minimum mean square error (MMSE) estimator and then transporting this estimate to high-quality images using a latent consistency flow-based model. Consequently, ELIR is more than 4x faster compared to the state-of-the-art diffusion and flow-based approaches. Moreover, ELIR is also more than 4x smaller, making it well-suited for deployment on resource-constrained edge devices. Comprehensive evaluations of various image restoration tasks show that ELIR achieves competitive results, effectively balancing distortion and perceptual quality metrics while offering improved efficiency in terms of memory and computation., Comment: 21 pages, 11 figures

Recent advances in efficient sequence modeling have introduced selective state-space layers, a key component of the Mamba architecture, which have demonstrated remarkable success in a wide range of NLP and vision tasks. While Mamba's empirical performance has matched or surpassed SoTA transformers on such diverse benchmarks, the theoretical foundations underlying its powerful representational capabilities remain less explored. In this work, we investigate the expressivity of selective state-space layers using multivariate polynomials, and prove that they surpass linear transformers in expressiveness. Consequently, our findings reveal that Mamba offers superior representational power over linear attention-based models for long sequences, while not sacrificing their generalization. Our theoretical insights are validated by a comprehensive set of empirical experiments on various datasets.

In $(t, n)$-threshold secret sharing, a secret $S$ is distributed among $n$ participants such that any subset of size $t$ can recover $S$, while any subset of size $t-1$ or fewer learns nothing about it. For information-theoretic secret sharing, it is known that the share size must be at least as large as the secret, i.e., $|S|$. When computational security is employed using cryptographic encryption with a secret key $K$, previous work has shown that the share size can be reduced to $\tfrac{|S|}{t} + |K|$. In this paper, we present a construction achieving a share size of $\tfrac{|S| + |K|}{t}$. Furthermore, we prove that, under reasonable assumptions on the encryption scheme -- namely, the non-compressibility of pseudorandom encryption and the non-redundancy of the secret key -- this share size is optimal.

As small Uncrewed Aircraft Systems (sUAS) increasingly operate in the national airspace, safety concerns arise due to a corresponding rise in reported airspace violations and incidents, highlighting the need for a safe mechanism for sUAS entry control to manage the potential overload. This paper presents work toward our aim of establishing automated, customized safety-claim support for managing on-entry requests from sUAS to enter controlled airspace. We describe our approach, Safety Case Software Product Line Engineering (SafeSPLE), which is a novel method to extend product-family techniques to on-entry safety cases. It begins with a hazard analysis and design of a safety case feature model defining key points in variation, followed by the creation of a parameterized safety case. We use these together to automate the generation of instances for specific sUAS. Finally we use a case study to demonstrate that the SafeSPLE method can be used to facilitate creation of safety cases for specific flights., Comment: Accepted at AIAA 2024

We propose SCONE ($\textbf{S}$calable, $\textbf{C}$ontextualized, $\textbf{O}$ffloaded, $\textbf{N}$-gram $\textbf{E}$mbedding), a method for extending input embedding layers to enhance language model performance as layer size scales. To avoid increased decoding costs, SCONE retains the original vocabulary while introducing embeddings for a set of frequent $n$-grams. These embeddings provide contextualized representation for each input token and are learned with a separate model during training. During inference, they are precomputed and stored in off-accelerator memory with minimal impact on inference speed. SCONE enables two new scaling strategies: increasing the number of cached $n$-gram embeddings and scaling the model used to learn them, all while maintaining fixed inference-time FLOPS. We show that scaling both aspects allows SCONE to outperform a 1.9B parameter baseline across diverse corpora, while using only half the inference-time FLOPS.

The rise of cyber-physical systems in safety-critical domains calls for robust risk-evaluation frameworks. Assurance cases, often required by regulatory bodies, are a structured approach to demonstrate that a system meets its safety requirements. However, assurance cases are fraught with challenges, such as incomplete evidence and gaps in reasoning, called defeaters, that can call into question the credibility and robustness of assurance cases. Identifying these defeaters increases confidence in the assurance case and can prevent catastrophic failures. The search for defeaters in an assurance case, however, is not structured, and there is a need to standardize defeater analysis. The software engineering community thus could benefit from having a reusable classification of real-world defeaters in software assurance cases. In this paper, we conducted a systematic study of literature from the past 20 years. Using open coding, we derived a taxonomy with seven broad categories, laying the groundwork for standardizing the analysis and management of defeaters in safety-critical systems. We provide our artifacts as open source for the community to use and build upon, thus establishing a common framework for understanding defeaters., Comment: ICSE 2025, Workshop on Multi-disciplinary, Open, and integRatEd Requirements Engineering

The cuneiform writing system served as the medium for transmitting knowledge in the ancient Near East for a period of over three thousand years. Cuneiform signs have a complex internal structure which is the subject of expert paleographic analysis, as variations in sign shapes bear witness to historical developments and transmission of writing and culture over time. However, prior automated techniques mostly treat sign types as categorical and do not explicitly model their highly varied internal configurations. In this work, we present an unsupervised approach for recovering the fine-grained internal configuration of cuneiform signs by leveraging powerful generative models and the appearance and structure of prototype font images as priors. Our approach, ProtoSnap, enforces structural consistency on matches found with deep image features to estimate the diverse configurations of cuneiform characters, snapping a skeleton-based template to photographed cuneiform signs. We provide a new benchmark of expert annotations and evaluate our method on this task. Our evaluation shows that our approach succeeds in aligning prototype skeletons to a wide variety of cuneiform signs. Moreover, we show that conditioning on structures produced by our method allows for generating synthetic data with correct structural configurations, significantly boosting the performance of cuneiform sign recognition beyond existing techniques, in particular over rare signs. Our code, data, and trained models are available at the project page: https://tau-vailab.github.io/ProtoSnap/, Comment: Accepted to ICLR 2025. Project page: https://tau-vailab.github.io/ProtoSnap/

The paper introduces a (universal) C*-algebra of continuous functions vanishing at infinity on the n-dimensional quantum complex space. To this end, the well-behaved Hilbert space representations of the defining relations are classified. Then these representations are realized by multiplication operators on an L2-space. The C*-algebra of continuous functions vanishing at infinity is defined by considering a *-algebra such that its classical counterpart separates the points of the n-dimensional complex space and by taking the operator norm closure of a universal representation of this algebra., Comment: published

We consider a Markov decision process (MDP) in which actions prescribed by the controller are executed by a separate actuator, which may behave adversarially. At each time step, the controller selects and transmits an action to the actuator; however, the actuator may deviate from the intended action to degrade the control reward. Given that the controller observes only the sequence of visited states, we investigate whether the actuator can covertly deviate from the controller's policy to minimize its reward without being detected. We establish conditions for covert adversarial behavior over an infinite time horizon and formulate an optimization problem to determine the optimal adversarial policy under these conditions. Additionally, we derive the asymptotic error exponents for detection in two scenarios: (1) a binary hypothesis testing framework, where the actuator either follows the prescribed policy or a known adversarial strategy, and (2) a composite hypothesis testing framework, where the actuator may employ any stationary policy. For the latter case, we also propose an optimization problem to maximize the adversary's performance.

This paper proposes a Right to AI, which asserts that individuals and communities should meaningfully participate in the development and governance of the AI systems that shape their lives. Motivated by the increasing deployment of AI in critical domains and inspired by Henri Lefebvre's concept of the Right to the City, we reconceptualize AI as a societal infrastructure, rather than merely a product of expert design. In this paper, we critically evaluate how generative agents, large-scale data extraction, and diverse cultural values bring new complexities to AI oversight. The paper proposes that grassroots participatory methodologies can mitigate biased outcomes and enhance social responsiveness. It asserts that data is socially produced and should be managed and owned collectively. Drawing on Sherry Arnstein's Ladder of Citizen Participation and analyzing nine case studies, the paper develops a four-tier model for the Right to AI that situates the current paradigm and envisions an aspirational future. It proposes recommendations for inclusive data ownership, transparent design processes, and stakeholder-driven oversight. We also discuss market-led and state-centric alternatives and argue that participatory approaches offer a better balance between technical efficiency and democratic legitimacy., Comment: 19 pages, 2 figures, 1 table

In this paper, we address the problem of secure distributed computation in scenarios where user data is not uniformly distributed, extending existing frameworks that assume uniformity, an assumption that is challenging to enforce in data for computation. Motivated by the pervasive reliance on single service providers for data storage and computation, we propose a privacy-preserving scheme that achieves information-theoretic security guarantees for computing polynomials over non-uniform data distributions. Our framework builds upon the concept of perfect subset privacy and employs linear hashing techniques to transform non-uniform data into approximately uniform distributions, enabling robust and secure computation. We derive leakage bounds and demonstrate that information leakage of any subset of user data to untrusted service providers, i.e., not only to colluding workers but also (and more importantly) to the admin, remains negligible under the proposed scheme., Comment: Parts of this work were submitted to ISIT 2025

A promising path to realizing higher superconducting transition temperatures $T_c$ is the strategic engineering of artificial heterostructures. For example, quantum materials exhibiting some but not all of the characteristics necessary for a robust superconducting state could, in principle, be coupled with other materials in a way that alleviates their intrinsic shortcomings. In this work, we add numerical support to the hypothesis that a strongly interacting superconductor weakened by phase fluctuations can boost its $T_c$ by hybridizing the system with a metal. Using determinant quantum Monte Carlo (DQMC), we simulate a two-dimensional bilayer composed of an attractive Hubbard model and a metallic layer in two regimes of the interaction strength $-|U|$. In the strongly interacting regime, we find that increasing the interlayer hybridization $t_\perp$ results in a nonmonotonic enhancement of $T_c$, with an optimal value comparable to the maximum $T_c$ observed in the single-layer attractive Hubbard model, confirming trends inferred from other approaches. In the intermediate coupling regime, when $-|U|$ is close to the value associated with the maximum $T_c$ of the single-layer model, increasing $t_\perp$ tends to decrease $T_c$, implying that the correlated layer was already optimally tuned. Importantly, we demonstrate that the mechanism behind these trends is related to enhancement in the superfluid stiffness, as was initially proposed by Kivelson [Physica B: Condensed Matter 318, 61 (2002)].

The fields of machine learning (ML) and cryptanalysis share an interestingly common objective of creating a function, based on a given set of inputs and outputs. However, the approaches and methods in doing so vary vastly between the two fields. In this paper, we explore integrating the knowledge from the ML domain to provide empirical evaluations of cryptosystems. Particularly, we utilize information theoretic metrics to perform ML-based distribution estimation. We propose two novel applications of ML algorithms that can be applied in a known plaintext setting to perform cryptanalysis on any cryptosystem. We use mutual information neural estimation to calculate a cryptosystem's mutual information leakage, and a binary cross entropy classification to model an indistinguishability under chosen plaintext attack (CPA). These algorithms can be readily applied in an audit setting to evaluate the robustness of a cryptosystem and the results can provide a useful empirical bound. We evaluate the efficacy of our methodologies by empirically analyzing several encryption schemes. Furthermore, we extend the analysis to novel network coding-based cryptosystems and provide other use cases for our algorithms. We show that our classification model correctly identifies the encryption schemes that are not IND-CPA secure, such as DES, RSA, and AES ECB, with high accuracy. It also identifies the faults in CPA-secure cryptosystems with faulty parameters, such a reduced counter version of AES-CTR. We also conclude that with our algorithms, in most cases a smaller-sized neural network using less computing power can identify vulnerabilities in cryptosystems, providing a quick check of the sanity of the cryptosystem and help to decide whether to spend more resources to deploy larger networks that are able to break the cryptosystem.

Resource-constrained edge deployments demand AI solutions that balance high performance with stringent compute, memory, and energy limitations. In this survey, we present a comprehensive overview of the primary strategies for accelerating deep learning models under such constraints. First, we examine model compression techniques-pruning, quantization, tensor decomposition, and knowledge distillation-that streamline large models into smaller, faster, and more efficient variants. Next, we explore Neural Architecture Search (NAS), a class of automated methods that discover architectures inherently optimized for particular tasks and hardware budgets. We then discuss compiler and deployment frameworks, such as TVM, TensorRT, and OpenVINO, which provide hardware-tailored optimizations at inference time. By integrating these three pillars into unified pipelines, practitioners can achieve multi-objective goals, including latency reduction, memory savings, and energy efficiency-all while maintaining competitive accuracy. We also highlight emerging frontiers in hierarchical NAS, neurosymbolic approaches, and advanced distillation tailored to large language models, underscoring open challenges like pre-training pruning for massive networks. Our survey offers practical insights, identifies current research gaps, and outlines promising directions for building scalable, platform-independent frameworks to accelerate deep learning models at the edge., Comment: 26 pages, 13 Figures

Solar energetic particle (SEP) events have been observed for decades in the interplanetary medium by spacecraft measuring the intensity of energetic ions and electrons. These intensities provide valuable information about particle acceleration, the effects of bulk plasma dynamics on particle transport, and the anisotropy of particle distributions. Since measured intensities are typically reported in narrow energy bins, it is common to re-bin intensities over a wider energy range to improve counting statistics. We investigate two methods for calculating intensities across multiple energy bins: a) \textit{re-binned intensity} (\(\overline{j}_{\rm linlin}\)), which is calculated by integrating the intensity over energy space and corresponds to the intensity at an effective energy that depends on the time-varying spectral index, and b) \textit{spectral binned intensity} (\(\overline{j}_{\rm loglog}\)), calculated by integrating the log-intensity in log-energy space, yielding the intensity at the log-centered energy that is independent of the spectral index and remains constant over time. We compare these methods using Parker Solar Probe (PSP) IS\(\odot\)IS measurements of energetic protons, and we prescribe criteria for selecting the appropriate method for different scenarios. Our results show that the re-binned intensity is consistently larger (up to a factor of 5) than the spectral binned intensity for two SEP events observed by PSP, although the time series of the two methods are strongly correlated. Overall, both measures are important for SEP spectral analysis, and the selection of the appropriate measure depends on whether a physical (spectral binned intensity) or a statistical (re-binned intensity) representation is needed for a given analysis., Comment: 17 pages, 9 Figures, Accepted for Publication in ApJS

Combining reinforcement learning with language grounding is challenging as the agent needs to explore the environment while simultaneously learning multiple language-conditioned tasks. To address this, we introduce a novel method: the compositionally-enabled reinforcement learning language agent (CERLLA). Our method reduces the sample complexity of tasks specified with language by leveraging compositional policy representations and a semantic parser trained using reinforcement learning and in-context learning. We evaluate our approach in an environment requiring function approximation and demonstrate compositional generalization to novel tasks. Our method significantly outperforms the previous best non-compositional baseline in terms of sample complexity on 162 tasks designed to test compositional generalization. Our model attains a higher success rate and learns in fewer steps than the non-compositional baseline. It reaches a success rate equal to an oracle policy's upper-bound performance of 92%. With the same number of environment steps, the baseline only reaches a success rate of 80%., Comment: TMLR 2024

We consider the problem where an active Decision-Maker (DM) is tasked to identify the true hypothesis using as few as possible observations while maintaining accuracy. The DM collects observations according to its determined actions and knows the distributions under each hypothesis. We propose a deterministic and adaptive multi-stage hypothesis-elimination strategy where the DM selects an action, applies it repeatedly, and discards hypotheses in light of its obtained observations. The DM selects actions based on maximal separation expressed by the distance between the parameter vectors of each distribution under each hypothesis. Close distributions can be clustered, simplifying the search and significantly reducing the number of required observations. Our algorithms achieve vanishing Average Bayes Risk (ABR) as the error probability approaches zero, i.e., the algorithm is asymptotically optimal. Furthermore, we show that the ABR is bounded when the number of hypotheses grows. Simulations are carried out to evaluate the algorithm's performance compared to another multi-stage hypothesis-elimination algorithm, where an improvement of several orders of magnitude in the mean number of observations required is observed., Comment: 7 pages, 2 figures

In this paper, we address the problem of Private Information Retrieval (PIR) over a public Additive White Gaussian Noise (AWGN) channel. In such a setup, the server's responses are visible to other servers. Thus, a curious server can listen to the other responses, compromising the user's privacy. Indeed, previous works on PIR over a shared medium assumed the servers cannot instantaneously listen to other responses. To address this gap, we present a novel randomized lattice -- PIR coding scheme that jointly codes for privacy, channel noise, and curious servers which may listen to other responses. We demonstrate that a positive PIR rate is achievable even in cases where the channel to the curious server is stronger than the channel to the user.

The adiabatic theorem in quantum mechanics implies that if a system is in a discrete eigenstate of a Hamiltonian and the Hamiltonian evolves in time arbitrarily slowly, the system will remain in the corresponding eigenstate of the evolved Hamiltonian. Understanding corrections to the adiabatic result that arise when the evolution of the Hamiltonian is slow -- but not arbitrarily slow -- has become increasingly important, especially since adiabatic evolution has been proposed as a method of state preparation in quantum computing. This paper identifies two regimes, an adiabatic regime in which corrections are generically small and can depend on details of the evolution throughout the path, and a hyperadiabatic regime in which the error is given by a form similar to an asymptotic expansion in the inverse of the evolution time with the coefficients depending principally on the behavior at the endpoints. However, the error in this hyperadiabatic regime is neither given by a true asymptotic series nor solely dependent on the endpoints: the coefficients combine the contributions from both endpoints, with relative phase factors that depend on the average spectral gaps along the trajectory, multiplied by the evolution time. The central result of this paper is to identify a quantity, referred to as the typical error, which is obtained by appropriately averaging the error over evolution times that are small compared to the evolution time itself. This typical error is characterized by an asymptotic series and depends solely on the endpoints of the evolution, remaining independent of the details of the intermediate evolution., Comment: 5 pages, 2 figures

Ap\'ery's remarkable discovery of rapidly converging continued fractions with small coefficients for $\zeta(2)$ and $\zeta(3)$ has led to a flurry of important activity in an incredible variety of different directions. Our purpose is to show that modifications of Ap\'ery's continued fractions can give interesting results including new rapidly convergent continued fractions for certain interesting constants., Comment: 14 pages

This paper introduces a novel method, Sample-efficient Probabilistic Detection using Extreme Value Theory (SPADE), which transforms a classifier into an abstaining classifier, offering provable protection against out-of-distribution and adversarial samples. The approach is based on a Generalized Extreme Value (GEV) model of the training distribution in the classifier's latent space, enabling the formal characterization of OOD samples. Interestingly, under mild assumptions, the GEV model also allows for formally characterizing adversarial samples. The abstaining classifier, which rejects samples based on their assessment by the GEV model, provably avoids OOD and adversarial samples. The empirical validation of the approach, conducted on various neural architectures (ResNet, VGG, and Vision Transformer) and medium and large-sized datasets (CIFAR-10, CIFAR-100, and ImageNet), demonstrates its frugality, stability, and efficiency compared to the state of the art., Comment: under review

The decoding of continuously spoken speech from neuronal activity has the potential to become an important clinical solution for paralyzed patients. Deep Learning Brain Computer Interfaces (BCIs) have recently successfully mapped neuronal activity to text contents in subjects who attempted to formulate speech. However, only small BCI datasets are available. In contrast, labeled data and pre-trained models for the closely related task of speech recognition from audio are widely available. One such model is Wav2Vec2 which has been trained in a self-supervised fashion to create meaningful representations of speech audio data. In this study, we show that patterns learned by Wav2Vec2 are transferable to brain data. Specifically, we replace its audio feature extractor with an untrained Brain Feature Extractor (BFE) model. We then execute full fine-tuning with pre-trained weights for Wav2Vec2, training ''from scratch'' without pre-trained weights as well as freezing a pre-trained Wav2Vec2 and training only the BFE each for 45 different BFE architectures. Across these experiments, the best run is from full fine-tuning with pre-trained weights, achieving a Character Error Rate (CER) of 18.54\%, outperforming the best training from scratch run by 20.46\% and that of frozen Wav2Vec2 training by 15.92\% percentage points. These results indicate that knowledge transfer from audio speech recognition to brain decoding is possible and significantly improves brain decoding performance for the same architectures. Related source code is available at https://github.com/tfiedlerdev/Wav2Vec2ForBrain., Comment: Paper was submitted to ICASSP 2025 but marginally rejected

Recent advancements in long-context language models (LCLMs) promise to transform Retrieval-Augmented Generation (RAG) by simplifying pipelines. With their expanded context windows, LCLMs can process entire knowledge bases and perform retrieval and reasoning directly -- a capability we define as In-Context Retrieval and Reasoning (ICR^2). However, existing benchmarks like LOFT often overestimate LCLM performance by providing overly simplified contexts. To address this, we introduce ICR^2, a benchmark that evaluates LCLMs in more realistic scenarios by including confounding passages retrieved with strong retrievers. We then propose three methods to enhance LCLM performance: (1) retrieve-then-generate fine-tuning, (2) retrieval-attention-probing, which uses attention heads to filter and de-noise long contexts during decoding, and (3) joint retrieval head training alongside the generation head. Our evaluation of five well-known LCLMs on LOFT and ICR^2 demonstrates significant gains with our best approach applied to Mistral-7B: +17 and +15 points by Exact Match on LOFT, and +13 and +2 points on ICR^2, compared to vanilla RAG and supervised fine-tuning, respectively. It even outperforms GPT-4-Turbo on most tasks despite being a much smaller model.

With the rising prevalence of cardiovascular and respiratory disorders and an aging global population, healthcare systems face increasing pressure to adopt efficient, non-contact vital sign monitoring (NCVSM) solutions. This study introduces a robust framework for multi-person localization and vital signs monitoring, using multiple-input-multiple-output frequency-modulated continuous wave radar, addressing challenges in real-world, cluttered environments. Two key contributions are presented. First, a custom hardware phantom was developed to simulate multi-person NCVSM scenarios, utilizing recorded thoracic impedance signals to replicate realistic cardiopulmonary dynamics. The phantom's design facilitates repeatable and rapid validation of radar systems and algorithms under diverse conditions to accelerate deployment in human monitoring. Second, aided by the phantom, we designed a robust algorithm for multi-person localization utilizing joint sparsity and cardiopulmonary properties, alongside harmonics-resilient dictionary-based vital signs estimation, to mitigate interfering respiration harmonics. Additionally, an adaptive signal refinement procedure is introduced to enhance the accuracy of continuous NCVSM by leveraging the continuity of the estimates. Performance was validated and compared to existing techniques through 12 phantom trials and 12 human trials, including both single- and multi-person scenarios, demonstrating superior localization and NCVSM performance. For example, in multi-person human trials, our method achieved average respiration rate estimation accuracies of 94.14%, 98.12%, and 98.69% within error thresholds of 2, 3, and 4 breaths per minute, respectively, and heart rate accuracies of 87.10%, 94.12%, and 95.54% within the same thresholds. These results highlight the potential of this framework for reliable multi-person NCVSM in healthcare and IoT applications.

We consider coresets for $k$-clustering problems, where the goal is to assign points to centers minimizing powers of distances. A popular example is the $k$-median objective $\sum_{p}\min_{c\in C}dist(p,C)$. Given a point set $P$, a coreset $\Omega$ is a small weighted subset that approximates the cost of $P$ for all candidate solutions $C$ up to a $(1\pm\varepsilon )$ multiplicative factor. In this paper, we give a sharp VC-dimension based analysis for coreset construction. As a consequence, we obtain improved $k$-median coreset bounds for the following metrics: Coresets of size $\tilde{O}\left(k\varepsilon^{-2}\right)$ for shortest path metrics in planar graphs, improving over the bounds $\tilde{O}\left(k\varepsilon^{-6}\right)$ by [Cohen-Addad, Saulpic, Schwiegelshohn, STOC'21] and $\tilde{O}\left(k^2\varepsilon^{-4}\right)$ by [Braverman, Jiang, Krauthgamer, Wu, SODA'21]. Coresets of size $\tilde{O}\left(kd\ell\varepsilon^{-2}\log m\right)$ for clustering $d$-dimensional polygonal curves of length at most $m$ with curves of length at most $\ell$ with respect to Frechet metrics, improving over the bounds $\tilde{O}\left(k^3d\ell\varepsilon^{-3}\log m\right)$ by [Braverman, Cohen-Addad, Jiang, Krauthgamer, Schwiegelshohn, Toftrup, and Wu, FOCS'22] and $\tilde{O}\left(k^2d\ell\varepsilon^{-2}\log m \log |P|\right)$ by [Conradi, Kolbe, Psarros, Rohde, SoCG'24].

Quantum correlations, like entanglement, represent the characteristic trait of quantum mechanics, and pose essential issues and challenges to the interpretation of this pillar of modern physics. Although quantum correlations are largely acknowledged as a major resource to achieve quantum advantage in many tasks of quantum technologies, their full quantitative description and the axiomatic basis underlying them are still under investigation. Previous works suggested that the origin of nonlocal correlations is grounded in principles capturing (from outside the quantum formalism) the essence of quantum uncertainty. In particular, the recently-introduced principle of Relativistic Independence gave rise to a new bound intertwining local and nonlocal correlations. Here we test such a bound by realizing together sequential and joint weak measurements on entangled photon pairs, allowing to simultaneously quantify both local and nonlocal correlations by measuring incompatible observables on the same quantum system without collapsing its state, a task typically forbidden in the traditional (projective) quantum measurement framework. Our results demonstrate the existence of a fundamental limit on the extent of quantum correlations, shedding light on the profound role of uncertainty in both enabling and balancing them., Comment: 23 pages, 3 figures

The nucleation and/or spreading of bubbles in water under tension (due to water evaporation) can be problematic for most plants along the ascending sap network from root to leaves, named xylem. Due to global warming, trees facing drought conditions are particularly threatened by the formation of such air embolisms, which spreads intermittently and hinder the flow of sap and could ultimately result in their demise. PDMS-based biomimetic leaves simulating evapotranspiration have demonstrated that, in a linear configuration, the existence of a slender constriction in the channel allows for the creation of intermittent embolism propagation (as an interaction between the elasticity of the biomimetic leaf (mainly the deformable ceiling of the microchannels) and the capillary forces at the air/water interfaces) \cite{Keiser2022}-\cite{keiser2024}. Here we use analog PDMS-based biomimetic leaves in 1d and 2d. To better explore the embolism spreading mechanism, we add to the setup an additional technique, allowing to measure directly the microchannel's ceiling deformation versus time, which corresponds to the pressure variations. We present here such a method that allows to have quantitative insights in the dynamics of embolism spreading. The coupling between channel deformations and the Laplace pressure threshold explains the observed elastocapillary dynamics.

For 3-manifolds with torus boundary, the bordered Heegaard Floer invariants of Lipshitz--Ozsv\'ath--Thurston have a geometric interpretation as immersed multi-curves with local systems in the punctured torus according to the work of Hanselman--Rasmussen--Watson. We consider morphisms between these immersed curve invariants and show that they compute certain cobordism maps. More precisely, we relate composition in the Fukaya category of immersed curves in the punctured torus to composition of morphisms between the bordered Floer invariants, which have interpretations in terms of certain cobordism maps. We make use of this formalism to obstruct smooth equivalences between 4-manifolds with boundary, and between surfaces with boundary in the 4-ball.

In blockchain systems, fair transaction ordering is crucial for a trusted and regulation-compliant economic ecosystem. Unlike traditional State Machine Replication (SMR) systems, which focus solely on liveness and safety, blockchain systems also require a fairness property. This paper examines these properties and aims to eliminate algorithmic bias in transaction ordering services. We build on the notion of equal opportunity. We characterize transactions in terms of relevant and irrelevant features, requiring that the order be determined solely by the relevant ones. Specifically, transactions with identical relevant features should have an equal chance of being ordered before one another. We extend this framework to define a property where the greater the distance in relevant features between transactions, the higher the probability of prioritizing one over the other. We reveal a surprising link between equal opportunity in SMR and Differential Privacy (DP), showing that any DP mechanism can be used to ensure fairness in SMR. This connection not only enhances our understanding of the interplay between privacy and fairness in distributed computing but also opens up new opportunities for designing fair distributed protocols using well-established DP techniques.

The study of complex adaptive systems, pioneered in physics, biology, and the social sciences, offers important lessons for AI governance. Contemporary AI systems and the environments in which they operate exhibit many of the properties characteristic of complex systems, including nonlinear growth patterns, emergent phenomena, and cascading effects that can lead to tail risks. Complexity theory can help illuminate the features of AI that pose central challenges for policymakers, such as feedback loops induced by training AI models on synthetic data and the interconnectedness between AI systems and critical infrastructure. Drawing on insights from other domains shaped by complex systems, including public health and climate change, we examine how efforts to govern AI are marked by deep uncertainty. To contend with this challenge, we propose a set of complexity-compatible principles concerning the timing and structure of AI governance, and the risk thresholds that should trigger regulatory intervention.

Vector graphics are essential in design, providing artists with a versatile medium for creating resolution-independent and highly editable visual content. Recent advancements in vision-language and diffusion models have fueled interest in text-to-vector graphics generation. However, existing approaches often suffer from over-parameterized outputs or treat the layered structure - a core feature of vector graphics - as a secondary goal, diminishing their practical use. Recognizing the importance of layered SVG representations, we propose NeuralSVG, an implicit neural representation for generating vector graphics from text prompts. Inspired by Neural Radiance Fields (NeRFs), NeuralSVG encodes the entire scene into the weights of a small MLP network, optimized using Score Distillation Sampling (SDS). To encourage a layered structure in the generated SVG, we introduce a dropout-based regularization technique that strengthens the standalone meaning of each shape. We additionally demonstrate that utilizing a neural representation provides an added benefit of inference-time control, enabling users to dynamically adapt the generated SVG based on user-provided inputs, all with a single learned representation. Through extensive qualitative and quantitative evaluations, we demonstrate that NeuralSVG outperforms existing methods in generating structured and flexible SVG., Comment: Project Page: https://sagipolaczek.github.io/NeuralSVG/

We determine a simple condition on a particular state graph of an alternating knot or link diagram that characterizes when the unoriented genus and crosscap number coincide, extending work of Adams and Kindred. Building on this same work and using continued fraction expansions, we provide a new formula for the unoriented genus of a 2-bridge knot or link. We use recursion to obtain exact formulas for the average unoriented genus $\overline{\Gamma}(c)$ and average crosscap number $\overline{\gamma}(c)$ of all 2-bridge knots with crossing number $c$, and in particular we show that $\displaystyle{\lim_{c\to\infty} \left(\frac{c}{3}+\frac{1}{9} - \overline{\Gamma}(c)\right) = \lim_{c\to\infty} \left(\frac{c}{3}+\frac{1}{9} - \overline{\gamma}(c)\right) = 0}$., Comment: 34 pages, 11 figures, 6 tables, comments welcome

On September 5, 2022, during Parker Solar Probe's (PSP) 13th encounter, a fast shock wave and a related solar energetic particle (SEP) event were observed as the spacecraft approached the perihelion of its orbit. Observations from the Integrated Science Investigation of the Sun (ISOIS) instrument suite show that SEPs arrived at the spacecraft with a significant delay from the onset of the parent solar eruption and that the first arriving SEPs exhibited an Inverse Velocity Dispersion (IVD) for energetic protons above $\sim$1~MeV. Utilizing data from multiple spacecraft we investigate the eruption dynamics and shock wave propagation. Our analysis includes 3D shock modeling and SEP transport simulations to examine the origins of this SEP event and explore the causes of the delayed SEP onset and the observed IVD. The data-driven SEP simulation reproduces the SEP event onset observed at PSP, its evolving energy spectrum and the IVD. This IVD is attributed to a relatively slow, ongoing particle acceleration process occurring at the flank of the expanding shock wave intercepted by PSP. This has significant implications for the role of shocks in the release of SEPs at widespread events and for methods used to infer the SEP release times. Furthermore, the match between the simulation and observations worsens when cross-field diffusion is considered, indicating that SEP diffusion had a minor effect on this event. These findings underscore the complexity of SEP events and emphasize the need for advanced modelling approaches to better understand the role of shock waves and other physical processes in SEP acceleration and release., Comment: 21 pages, 11 figures, Accepted in The Astrophysical Journal

The Northwestern University Freight Rail Infrastructure and Energy Network Decarbonization (NUFRIEND) framework is a comprehensive industry-oriented tool for simulating the deployment of new energy technologies including biofuels, e-fuels, battery-electric, and hydrogen locomotives. By classifying fuel types into two categories based on deployment requirements, the associated optimal charging/fueling facility location and sizing problem are solved with a five-step framework. Life cycle analyses (LCA) and techno-economic analyses (TEA) are used to estimate carbon reduction, capital investments, cost of carbon reduction, and operational impacts, enabling sensitivity analysis with operational and technological parameters. The framework is illustrated on lower-carbon drop-in fuels as well as battery-electric technology deployments for US Eastern and Western Class I railroad networks. Drop-in fuel deployments are modeled as admixtures with diesel in existing locomotives, while battery-electric deployments are shown for varying technology penetration levels and locomotive ranges. When mixed in a 50 percent ratio with diesel, results show biodiesel's capacity to reduce emissions at 36 percent with a cost of 0.13 USD per kilogram of CO2 reduced, while e-fuels offer a 50 percent emissions reduction potential at a cost of 0.22 USD per kilogram of CO2 reduced. Battery-electric results for 50 percent deployment over all ton-miles highlight the value of future innovations in battery energy densities as scenarios assuming 800-mile range locomotives show an estimated emissions reduction of 46 percent with a cost of 0.06 USD per kilogram of CO2 reduced, compared to 16 percent emissions reduction at a cost of 0.11 USD per kilogram of CO2 reduced for 400-mile range locomotives., Comment: 29 pages, 17 figures. This is the accepted version of a work that was published in Transportation Research Record

We introduce a method for composing object-level visual prompts within a text-to-image diffusion model. Our approach addresses the task of generating semantically coherent compositions across diverse scenes and styles, similar to the versatility and expressiveness offered by text prompts. A key challenge in this task is to preserve the identity of the objects depicted in the input visual prompts, while also generating diverse compositions across different images. To address this challenge, we introduce a new KV-mixed cross-attention mechanism, in which keys and values are learned from distinct visual representations. The keys are derived from an encoder with a small bottleneck for layout control, whereas the values come from a larger bottleneck encoder that captures fine-grained appearance details. By mixing keys and values from these complementary sources, our model preserves the identity of the visual prompts while supporting flexible variations in object arrangement, pose, and composition. During inference, we further propose object-level compositional guidance to improve the method's identity preservation and layout correctness. Results show that our technique produces diverse scene compositions that preserve the unique characteristics of each visual prompt, expanding the creative potential of text-to-image generation., Comment: Project: https://snap-research.github.io/visual-composer/

Personalizing text-to-image models to generate images of specific subjects across diverse scenes and styles is a rapidly advancing field. Current approaches often face challenges in maintaining a balance between identity preservation and alignment with the input text prompt. Some methods rely on a single textual token to represent a subject, which limits expressiveness, while others employ richer representations but disrupt the model's prior, diminishing prompt alignment. In this work, we introduce Nested Attention, a novel mechanism that injects a rich and expressive image representation into the model's existing cross-attention layers. Our key idea is to generate query-dependent subject values, derived from nested attention layers that learn to select relevant subject features for each region in the generated image. We integrate these nested layers into an encoder-based personalization method, and show that they enable high identity preservation while adhering to input text prompts. Our approach is general and can be trained on various domains. Additionally, its prior preservation allows us to combine multiple personalized subjects from different domains in a single image., Comment: Project page at https://snap-research.github.io/NestedAttention/

Epidemic spreading over populations networks has been an important subject of research for several decades, and especially during the Covid-19 pandemic. Most epidemic outbreaks are likely to create multiple mutations during their spreading over the population. In this paper, we study the evolution of a pathogen which can mutate continuously during the epidemic spreading. We consider pathogens whose mutating parameter is the mortality mean-time, and study the evolution of this parameter over the spreading process. We use analytical methods to compute the dynamic equation of the epidemic and the conditions for it to spread. We also use numerical simulations to study the pathogen flow in this case, and to understand the mutation phenomena. We show that the natural selection leads to less violent pathogens becoming predominant in the population. We discuss a wide range of network structures and show how different effects are manifested in each case.

This work proposes a gradient-based method to design bone implants using triply-periodic minimal surfaces (TPMS) of spatially varying thickness to maximize bone in-growth. Bone growth into the implant is estimated using a finite element based mechanobiological model considering the magnitude and frequency of in vivo loads, as well as the density distribution of the surrounding bone. The wall thicknesses of the implant unit cells are determined via linear interpolation of the thicknesses over a user defined grid of control points, avoiding mesh dependency and providing control over the sensitivity computation costs. The TPMS structure is modeled as a homogenized material to reduce computational cost. Local properties of the implant are determined at run-time on an element-by-element basis using a pre-constructed surrogate model of the TPMS's physical and geometric properties as a function of the local wall thickness and the density of in-grown bone. Design sensitivities of the bone growth within the implant are computed using the direct sensitivity method. The methodology is demonstrated on a cementless hip, optimizing the implant for bone growth subject to wall thickness constraints to ensure manufacturability and allow cell infiltration., Comment: 25 pages, 12 figures, submitted to Structural and Multidisciplinary Optimization

Infectious diseases remain a critical global health challenge, and the integration of standardized ontologies plays a vital role in managing related data. The Infectious Disease Ontology (IDO) and its extensions, such as the Coronavirus Infectious Disease Ontology (CIDO), are essential for organizing and disseminating information related to infectious diseases. The COVID-19 pandemic highlighted the need for updating IDO and its virus-specific extensions. There is an additional need to update IDO extensions specific to bacteria, fungus, and parasite infectious diseases. We adopt the "hub and spoke" methodology to generate pathogen-specific extensions of IDO: Virus Infectious Disease Ontology (VIDO), Bacteria Infectious Disease Ontology (BIDO), Mycosis Infectious Disease Ontology (MIDO), and Parasite Infectious Disease Ontology (PIDO). The creation of pathogen-specific reference ontologies advances modularization and reusability of infectious disease data within the IDO ecosystem. Future work will focus on further refining these ontologies, creating new extensions, and developing application ontologies based on them, in line with ongoing efforts to standardize biological and biomedical terminologies for improved data sharing and analysis., Comment: 25 pages

To support planning of alternative fuel technology (e.g., battery-electric locomotives) deployment for decarbonizing non-electrified freight rail, we develop a convex optimization formulation with a closed-form solution to determine the optimal number of energy storage tender cars in a train. The formulation shares a similar structure to an Economic Order Quantity (EOQ) model. For given market characteristics, cost forecasts, and technology parameters, our model captures the trade-offs between inventory carrying costs associated with trip times (including delays due to charging/refueling) and ordering costs associated with train dispatch and operation (energy, amortized equipment, and labor costs). To illustrate the framework, we find the optimal number of battery-electric energy tender cars in 22,501 freight markets (origin-destination pairs and commodities) for U.S. Class I railroads. The results display heterogeneity in optimal configurations with lighter, yet more time-sensitive shipments (e.g., intermodal) utilizing more battery tender cars. For heavier commodities (e.g., coal) with lower holding costs, single battery tender car configurations are generally optimal. The results also show that the optimal train configurations are sensitive to delays associated with recharging or swapping tender cars., Comment: 42 pages, 19 figures. This is the accepted version of a work that was published in Transportation Research Part E: Logistics and Transportation Review

We address the problem of searching for a change point in an anomalous process among a finite set of M processes. Specifically, we address a composite hypothesis model in which each process generates measurements following a common distribution with an unknown parameter (vector). This parameter belongs to either a normal or abnormal space depending on the current state of the process. Before the change point, all processes, including the anomalous one, are in a normal state; after the change point, the anomalous process transitions to an abnormal state. Our goal is to design a sequential search strategy that minimizes the Bayes risk by balancing sample complexity and detection accuracy. We propose a deterministic search algorithm with the following notable properties. First, we analytically demonstrate that when the distributions of both normal and abnormal processes are unknown, the algorithm is asymptotically optimal in minimizing the Bayes risk as the error probability approaches zero. In the second setting, where the parameter under the null hypothesis is known, the algorithm achieves asymptotic optimality with improved detection time based on the true normal state. Simulation results are presented to validate the theoretical findings., Comment: 13 pages, 6 figures

Balancing content fidelity and artistic style is a pivotal challenge in image generation. While traditional style transfer methods and modern Denoising Diffusion Probabilistic Models (DDPMs) strive to achieve this balance, they often struggle to do so without sacrificing either style, content, or sometimes both. This work addresses this challenge by analyzing the ability of DDPMs to maintain content and style equilibrium. We introduce a novel method to identify sensitivities within the DDPM attention layers, identifying specific layers that correspond to different stylistic aspects. By directing conditional inputs only to these sensitive layers, our approach enables fine-grained control over style and content, significantly reducing issues arising from over-constrained inputs. Our findings demonstrate that this method enhances recent stylization techniques by better aligning style and content, ultimately improving the quality of generated visual content.