arXiv subjects

Artificial intelligence

Search indexed arXiv papers on artificial intelligence and machine learning, including cs.AI metadata. Follow the original manuscripts for methods, experiments and version history.

Follow this topic with RSS

At least 163 records · Page 9Linked to original sources

arXiv2026

PHOTON: Hierarchical Autoregressive Modeling for Lightspeed and Memory-Efficient Language Generation

Transformers operate as horizontal token-by-token scanners; at each generation step, attending to an ever-growing sequence of token-level states. This access pattern increases prefill latency and makes long-context decoding more memory-bound, as KV-cache reads and writes dominate inference time over arithmetic operations. We propose Parallel Hierarchical Operation for TOp-down Networks (PHOTON), a hierarchical autoregressive model that replaces horizontal scanning with vertical, multi-resolution context scanning. PHOTON maintains a hierarchy of latent streams: a bottom-up encoder compresses tokens into low-rate contextual states, while lightweight top-down decoders reconstruct fine-grained token representations in parallel. We further introduce recursive generation that updates only the coarsest latent stream and eliminates bottom-up re-encoding. Experimental results show that PHOTON is superior to competitive Transformer-based language models regarding the throughput-quality trade-off, providing advantages in long-context and multi-query tasks. In particular, this reduces decode-time KV-cache traffic, yielding up to $10^{3}\times$ higher throughput per unit memory.

cs.LG↗

arXiv2026

FEM-Bench: A Structured Scientific Reasoning Benchmark for Evaluating Code-Generating LLMs

As LLMs advance their reasoning capabilities about the physical world, the absence of rigorous benchmarks for evaluating their ability to generate scientifically valid physical models has become a critical gap. Computational mechanics, which develops and applies mathematical models and numerical methods to predict the behavior of physical systems under forces, deformation, and constraints, provides an ideal foundation for structured scientific reasoning evaluation. Problems follow clear mathematical structure, enforce strict physical and numerical constraints, and support objective verification. The discipline requires constructing explicit models of physical systems and reasoning about geometry, spatial relationships, and material behavior, connecting directly to emerging AI goals in physical reasoning and world modeling. We introduce FEM-Bench, a computational mechanics benchmark designed to evaluate the ability of LLMs to generate correct finite element method (FEM) and related code. FEM-Bench 2025 contains a suite of introductory but nontrivial tasks aligned with material from a first graduate course on computational mechanics. These tasks capture essential numerical and physical modeling challenges while representing only a small fraction of the complexity present in the discipline. Despite their simplicity, state-of-the-art LLMs do not reliably solve all of them. In a five attempt run, the best performing model at function writing, Gemini 3 Pro, completed 30/33 tasks at least once and 26/33 tasks all five times. The best performing model at unit test writing, GPT-5, had an Average Joint Success Rate of 73.8%. Other popular models showed broad performance variation. FEM-Bench establishes a structured foundation for evaluating AI-generated scientific code, and future iterations will incorporate increasingly sophisticated tasks to track progress as models evolve.

cs.LG↗

arXiv2026

Generalization of RLVR Using Causal Reasoning as a Testbed

Reinforcement learning with verifiable rewards (RLVR) has emerged as a promising paradigm for post-training large language models (LLMs) on complex reasoning tasks. Yet, the conditions under which RLVR yields robust generalization remain underexplored. This paper provides an empirical study of RLVR generalization in the setting of probabilistic inference over causal graphical models. This setting offers two natural axes along which to examine generalization: (i) the level of the probabilistic query -- associational, interventional, or counterfactual -- and (ii) the structural complexity of the query, measured by the size of its relevant subgraph. We construct a dataset of causal graphs and queries spanning these difficulty axes and fine-tune Qwen-2.5-Instruct models using RLVR or supervised fine-tuning (SFT). We vary both the model scale (3B-32B) and the query level included in training. We find that RLVR yields stronger within-level and across-level generalization than SFT, but only for specific combinations of model size and training query level. Further analysis shows that RLVR's effectiveness depends on the model's initial reasoning competence. With sufficient initial competence, RLVR improves an LLM's marginalization strategy and reduces errors in intermediate probability calculations, producing substantial accuracy gains, particularly on more complex queries. These results show that RLVR can improve specific causal reasoning subskills, with its benefits emerging only when the model has sufficient initial competence. Our code and data is available at https://github.com/zhichul/rlcausal.

cs.LG↗

arXiv2026

Safety Alignment of LMs via Non-cooperative Games

Ensuring the safety of language models (LMs) while maintaining their usefulness remains a critical challenge in AI alignment. Current approaches rely on sequential adversarial training: generating adversarial prompts and fine-tuning LMs to defend against them. We introduce a different paradigm: framing safety alignment as a non-zero-sum game between an Attacker LM and a Defender LM trained jointly via online reinforcement learning. Each LM continuously adapts to the other's evolving strategies, driving iterative improvement. Our method uses a preference-based reward signal derived from pairwise comparisons instead of point-wise scores, providing more robust supervision and potentially reducing reward hacking. Our RL recipe, AdvGame, shifts the Pareto frontier of safety and utility, yielding a Defender LM that is simultaneously more helpful and more resilient to adversarial attacks. In addition, the resulting Attacker LM converges into a strong, general-purpose red-teaming agent that can be directly deployed to probe arbitrary target models. Code at github.com/facebookresearch/advgame.

cs.AI↗

arXiv2026

MediEval: A Unified Medical Benchmark for Patient-Contextual and Knowledge-Grounded Reasoning in LLMs

Large Language Models (LLMs) are increasingly applied to medicine, yet their adoption is limited by concerns over reliability and safety. Existing evaluations either test factual medical knowledge in isolation or assess patient-level reasoning without verifying correctness, leaving a critical gap. We introduce MediEval, a benchmark that links MIMIC-IV electronic health records (EHRs) to a unified knowledge base built from UMLS and other biomedical vocabularies. MediEval generates diverse factual and counterfactual medical statements within real patient contexts, enabling systematic evaluation across a 4-quadrant framework that jointly considers knowledge grounding and contextual consistency. Using this framework, we identify critical failure modes, including hallucinated support and truth inversion, that current proprietary, open-source, and domain-specific LLMs frequently exhibit. To address these risks, we propose Counterfactual Risk-Aware Fine-tuning (CoRFu), a DPO-based method with an asymmetric penalty targeting unsafe confusions. CoRFu improves by +16.4 macro-F1 points over the base model and eliminates truth inversion errors, demonstrating both higher accuracy and substantially greater safety.

cs.CL↗

arXiv2026

Context-Sensitive Abstractions for Reinforcement Learning with Parameterized Actions

Real-world sequential decision-making often involves parameterized action spaces that require both, decisions regarding discrete actions and decisions about continuous action parameters governing how an action is executed. Existing approaches exhibit severe limitations in this setting -- planning methods demand hand-crafted action models, and standard reinforcement learning (RL) algorithms are designed for either discrete or continuous actions but not both, and the few RL methods that handle parameterized actions typically rely on domain-specific engineering and fail to exploit the latent structure of these spaces. This paper extends the scope of RL algorithms to long-horizon, sparse-reward settings with parameterized actions by enabling agents to autonomously learn both state and action abstractions online. We introduce algorithms that progressively refine these abstractions during learning, increasing fine-grained detail in the critical regions of the state-action space where greater resolution improves performance. Across several continuous-state, parameterized-action domains, our abstraction-driven approach enables TD($λ$) to achieve markedly higher sample efficiency than state-of-the-art baselines.

cs.AI↗

arXiv2026

MAR:Multi-Agent Reflexion Improves Reasoning Abilities in LLMs

LLMs have shown the capacity to improve their performance on reasoning tasks through reflecting on their mistakes, and acting with these reflections in mind. However, continual reflections of the same LLM onto itself exhibit degeneration of thought, where the LLM continues to repeat the same errors again and again even with the knowledge that its wrong. To address this problem, we instead introduce multi-agent with multi-persona debators as the method to generate reflections. Through out extensive experimentation, we've found that the leads to better diversity of in the reflections generated by the llm agent. We demonstrate an accuracy of 47% EM HotPot QA (question answering) and 82.7% on HumanEval (programming), both performances surpassing reflection with a single llm.

cs.AI↗

arXiv2026

Memory-Efficient Acceleration of Block Low-Rank Foundation Models on Resource Constrained GPUs

Recent advances in transformer-based foundation models have made them the default choice for many tasks, but their rapidly growing size makes fitting a full model on a single GPU increasingly difficult and their computational cost prohibitive. Block low-rank (BLR) compression techniques address this challenge by learning compact representations of weight matrices. While traditional low-rank (LR) methods often incur sharp accuracy drops, BLR approaches such as Monarch and BLAST can better capture the underlying structure, thus preserving accuracy while reducing computations and memory footprints. In this work, we use roofline analysis to show that, although BLR methods achieve theoretical savings and practical speedups for single-token inference, multi-token inference often becomes memory-bound in practice, increasing latency despite compiler-level optimizations in PyTorch. To address this, we introduce custom Triton kernels with partial fusion and memory layout optimizations for both Monarch and BLAST. On memory-constrained NVIDIA GPUs such as Jetson Orin Nano and A40, our kernels deliver up to $3.76\times$ speedups and $3\times$ model size compression over PyTorch dense baselines using CUDA backend and compiler-level optimizations, while supporting various models including Llama-7/1B, GPT2-S, DiT-XL/2, and ViT-B. Our code is available at https://github.com/pabillam/mem-efficient-blr.

cs.LG↗

arXiv2026

DiEC: Diffusion Embedded Clustering

Deep clustering methods typically rely on a single, well-defined representation for clustering. In contrast, pretrained diffusion models provide abundant and diverse multi-scale representations across network layers and noise timesteps. However, a key challenge is how to efficiently identify the most clustering-friendly representation in the layer*timestep space. To address this issue, we propose Diffusion Embedded Clustering (DiEC), an unsupervised framework that performs clustering by leveraging optimal intermediate representations from pretrained diffusion models. DiEC systematically evaluates the clusterability of representations along the trajectory of network depth and noise timesteps. Meanwhile, an unsupervised search strategy is designed for recognizing the Clustering-optimal Layer (COL) and Clustering-optimal Timestep (COT) in the layer*timestep space of pretrained diffusion models, aiming to promote clustering performance and reduce computational overhead. DiEC is fine-tuned primarily with a structure-preserving DEC-style KL-divergence objective at the fixed COL + COT, together with a random-timestep diffusion denoising objective to maintain the generative capability of the pretrained model. Without relying on augmentation-based consistency constraints or contrastive learning, DiEC achieves excellent clustering performance across multiple benchmark datasets.

cs.LG↗

arXiv2016

Symbols of a cosmic order

The world runs on communicated sequences of symbols, e.g. numerals. Examining both engineered and natural communications networks reveals an unsuspected order that depends on contact with an unpredictable entity. This order has three roots. The first is a proof within quantum theory that no evidence can ever determine its explanation, so that an agent choosing an explanation must do so unpredictably. The second root is the showing that clocks that step computers do not "tell time" but serve as self-adjusting symbol-handling agents that regulate "logically synchronized" motion in response to unpredictable disturbances. Such a clock-agent has a certain independence as well as the capacity to communicate via unpredictable symbols with other clock-agents and to adjust its own tick rate in response to that communication. The third root is the noticing of unpredictable symbol exchange in natural systems, including the transmission of symbols found in molecular biology. We introduce a symbol-handling agent as a role played in some cases by a person, for example a physicist who chooses an explanation of given experimental outcomes, and in other cases by some other biological entity, and in still other cases by an inanimate device, such as a computer-based detector used in physical measurements. While we forbear to try to explain the propensity of agents at all levels from cells to civilizations to form and operate networks of logically synchronized symbol-handling agents, we point to this propensity as an overlooked cosmic order, an order structured by the unpredictability ensuing from the proof. Appreciating the cosmic order leads to a conception of agency that replaces volition by unpredictability and re-conceives the notion of objectivity in a way that makes a place for agency in the world as described by physics. Some specific implications for physics are outlined.

physics.hist-ph↗

arXiv2016

Tractability and Decompositions of Global Cost Functions

Enforcing local consistencies in cost function networks is performed by applying so-called Equivalent Preserving Transformations (EPTs) to the cost functions. As EPTs transform the cost functions, they may break the property that was making local consistency enforcement tractable on a global cost function. A global cost function is called tractable projection-safe when applying an EPT to it is tractable and does not break the tractability property. In this paper, we prove that depending on the size r of the smallest scopes used for performing EPTs, the tractability of global cost functions can be preserved (r = 0) or destroyed (r > 1). When r = 1, the answer is indefinite. We show that on a large family of cost functions, EPTs can be computed via dynamic programming-based algorithms, leading to tractable projection-safety. We also show that when a global cost function can be decomposed into a Berge acyclic network of bounded arity cost functions, soft local consistencies such as soft Directed or Virtual Arc Consistency can directly emulate dynamic programming. These different approaches to decomposable cost functions are then embedded in a solver for extensive experiments that confirm the feasibility and efficiency of our proposal.

cs.AI↗

arXiv2016

A Stochastic Temporal Model of Polyphonic MIDI Performance with Ornaments

We study indeterminacies in realization of ornaments and how they can be incorporated in a stochastic performance model applicable for music information processing such as score-performance matching. We point out the importance of temporal information, and propose a hidden Markov model which describes it explicitly and represents ornaments with several state types. Following a review of the indeterminacies, they are carefully incorporated into the model through its topology and parameters, and the state construction for quite general polyphonic scores is explained in detail. By analyzing piano performance data, we find significant overlaps in inter-onset-interval distributions of chordal notes, ornaments, and inter-chord events, and the data is used to determine details of the model. The model is applied for score following and offline score-performance matching, yielding highly accurate matching for performances with many ornaments and relatively frequent errors, repeats, and skips.

cs.AI↗

arXiv2014

A Novel Method for Comparative Analysis of DNA Sequences by Ramanujan-Fourier Transform

Alignment-free sequence analysis approaches provide important alternatives over multiple sequence alignment (MSA) in biological sequence analysis because alignment-free approaches have low computation complexity and are not dependent on high level of sequence identity, however, most of the existing alignment-free methods do not employ true full information content of sequences and thus can not accurately reveal similarities and differences among DNA sequences. We present a novel alignment-free computational method for sequence analysis based on Ramanujan-Fourier transform (RFT), in which complete information of DNA sequences is retained. We represent DNA sequences as four binary indicator sequences and apply RFT on the indicator sequences to convert them into frequency domain. The Euclidean distance of the complete RFT coefficients of DNA sequences are used as similarity measure. To address the different lengths in Euclidean space of RFT coefficients, we pad zeros to short DNA binary sequences so that the binary sequences equal the longest length in the comparison sequence data. Thus, the DNA sequences are compared in the same dimensional frequency space without information loss. We demonstrate the usefulness of the proposed method by presenting experimental results on hierarchical clustering of genes and genomes. The proposed method opens a new channel to biological sequence analysis, classification, and structural module identification.

cs.CE↗

arXiv2020

Synthetic Dataset Generation with Itemset-Based Generative Models

This paper proposes three different data generators, tailored to transactional datasets, based on existing itemset-based generative models. All these generators are intuitive and easy to implement and show satisfactory performance. The quality of each generator is assessed by means of three different methods that capture how well the original dataset structure is preserved.

cs.DB↗

arXiv2016

Leveraging Unstructured Data to Detect Emerging Reliability Issues

Unstructured data refers to information that does not have a predefined data model or is not organized in a pre-defined manner. Loosely speaking, unstructured data refers to text data that is generated by humans. In after-sales service businesses, there are two main sources of unstructured data: customer complaints, which generally describe symptoms, and technician comments, which outline diagnostics and treatment information. A legitimate customer complaint can eventually be tracked to a failure or a claim. However, there is a delay between the time of a customer complaint and the time of a failure or a claim. A proactive strategy aimed at analyzing customer complaints for symptoms can help service providers detect reliability problems in advance and initiate corrective actions such as recalls. This paper introduces essential text mining concepts in the context of reliability analysis and a method to detect emerging reliability issues. The application of the method is illustrated using a case study.

cs.AI↗

arXiv2016

Label Tree Embeddings for Acoustic Scene Classification

We present in this paper an efficient approach for acoustic scene classification by exploring the structure of class labels. Given a set of class labels, a category taxonomy is automatically learned by collectively optimizing a clustering of the labels into multiple meta-classes in a tree structure. An acoustic scene instance is then embedded into a low-dimensional feature representation which consists of the likelihoods that it belongs to the meta-classes. We demonstrate state-of-the-art results on two different datasets for the acoustic scene classification task, including the DCASE 2013 and LITIS Rouen datasets.

cs.MM↗

arXiv2016

Learning Abstract Classes using Deep Learning

Humans are generally good at learning abstract concepts about objects and scenes (e.g.\ spatial orientation, relative sizes, etc.). Over the last years convolutional neural networks have achieved almost human performance in recognizing concrete classes (i.e.\ specific object categories). This paper tests the performance of a current CNN (GoogLeNet) on the task of differentiating between abstract classes which are trivially differentiable for humans. We trained and tested the CNN on the two abstract classes of horizontal and vertical orientation and determined how well the network is able to transfer the learned classes to other, previously unseen objects.

cs.CV↗

arXiv2016

Interacting Conceptual Spaces

We propose applying the categorical compositional scheme of [6] to conceptual space models of cognition. In order to do this we introduce the category of convex relations as a new setting for categorical compositional semantics, emphasizing the convex structure important to conceptual space applications. We show how conceptual spaces for composite types such as adjectives and verbs can be constructed. We illustrate this new model on detailed examples.

cs.AI↗