Machine Learning · Cross-listed from cond-mat.mtrl-sci
AQVolt26: High-Temperature r$^2$SCAN Halide Dataset for Universal ML Potentials and Solid-State Batteries
Jiyoon Kim, Chuhong Wang, Aayush R. Singh et al. (+6 more)
The demand for safe, high-energy-density batteries has spotlighted halide solid-state electrolytes, which offer the potential for enhanced ionic mobility, electrochemical stability, and interfacial deformability. Accelerating their discovery requires extensive molecular dynamics, which has been increasingly enabled by universal machine learning interatomic potentials trained on foundational
datasets. However, the dynamic softness of halides poses a stringent test of whether general-purpose models can reliably replace first-principles calculations under the highly distorted, elevated-temperature regimes necessary to probe ion transport. Here, we present AQVolt26, a
dataset of 322,656 r$^2$SCAN single-point calculations for lithium halides, generated via high-temperature configurational sampling across $\sim$5K structures. We demonstrate that foundational
datasets provide a strong baseline for stable halide chemistries and transfer local forces well, however absolute energy predictions degrade in distorted higher-temperature regimes. Co-training with AQVolt26 resolves this blind spot. Furthermore, incorporating Materials Project relaxation data improves near-
equilibrium performance but degrades extreme-strain robustness without enhancing high-temperature force accuracy. These results demonstrate that domain-specific configurational sampling is essential for the reliable dynamic screening of halide electrolytes. Furthermore, our findings suggest that while foundational models provide a robust base, they are most effective for dynamically soft solid-state chemistries when augmented with targeted, high-temperature data. Finally, we show that near-
equilibrium relaxation data serves as a task-specific complement rather than a universally beneficial addition.
Read the full abstract ▾
All authors Jiyoon Kim, Chuhong Wang, Aayush R. Singh, Tyler Sours, Shivang Agarwal, AJ Nish, Paul Abruzzo, Ang Xiao, Omar Allam
Machine Learning · Cross-listed from physics.data-an
Neural posterior estimation for scalable and accurate inverse parameter inference in Li-ion batteries
Malik Hassanaly, Corey R. Randall, Peter J. Weddle et al. (+4 more)
Diagnosing the internal state of Li-ion batteries is critical for battery research, operation of real-world systems, and prognostic evaluation of remaining lifetime. By using physics-based models to perform probabilistic parameter estimation via
Bayesian calibration, diagnostics can account for the uncertainty due to model fitness, data noise, and the observability of any given parameter. However,
Bayesian calibration in Li-ion batteries using electrochemical data is computationally intensive even when using a fast surrogate in place of physics-based models, requiring many thousands of model evaluations. A fully amortized alternative is neural posterior estimation (NPE). NPE shifts the computational burden from the parameter estimation step to data generation and model training, reducing the parameter estimation time from minutes to milliseconds, enabling real-time applications. The present work shows that NPE calibrates parameters equally or more accurately than
Bayesian calibration, and we demonstrate that the higher computational costs for data generation are tractable even in high-dimensional cases (ranging from 6 to 27 estimated parameters), but the NPE method can lead to higher voltage prediction errors. The NPE method also offers several interpretability advantages over
Bayesian calibration, such as local parameter sensitivity to specific regions of the voltage curve. The NPE method is demonstrated using an experimental fast charge
dataset, with parameter estimates validated against measurements of loss of lithium inventory and loss of active material. The implementation is made available in a companion repository (https://github.com/NatLabRockies/BatFIT).
Read the full abstract ▾
All authors Malik Hassanaly, Corey R. Randall, Peter J. Weddle, Paul J. Gasper, Conlain Kelly, Tanvir R. Tanim, Kandler Smith
Machine Learning
Re-analysis of the Human Transcription Factor Atlas Recovers TF-Specific Signatures from Pooled Single-Cell Screens with Missing Controls
Arka Jain and Umesh Sharma
Public pooled single-
cell perturbation atlases are valuable resources for studying transcription factor (TF) function, but downstream re-analysis can be limited by incomplete deposited metadata and missing internal controls. Here we re-analyze the human TF Atlas
dataset (GSE216481), a MORF-based pooled overexpression screen spanning 3,550 TF open reading frames and 254,519
cells, with a reproducible pipeline for quality control, MORF barcode demultiplexing, per-TF differential expression, and functional enrichment. From 77,018
cells in the pooled screen, we assign 60,997 (79.2%) to 87 TF identities. Because the deposited barcode mapping lacks the GFP and mCherry negative controls present in the original library, we use embryoid body (EB)
cells as an external baseline and remove shared batch/transduction artifacts by background subtraction. This strategy recovers TF-specific signatures for 59 of 61 testable TFs, compared with 27 detected by one-vs-rest alone, showing that robust TF-level signal can be rescued despite missing intra-pool controls. HOPX, MAZ, PAX6, FOS, and FEZF2 emerge as the strongest transcriptional remodelers, while per-TF enrichment links FEZF2 to regulation of differentiation, EGR1 to Hippo and cardiac programs, FOS to focal adhesion, and NFIC to collagen biosynthesis. Condition-level analyses reveal convergent Wnt, neurogenic, EMT, and Hippo signatures, and Harmony indicates minimal confounding batch effects across pooled replicates. Our per-TF effect sizes significantly agree with Joung et al.'s published rankings (Spearman $\rho = -0.316$, $p = 0.013$; negative because lower rank indicates stronger effect). Together, these results show that the deposited TF Atlas data can support validated TF-specific transcriptional and pathway analyses when paired with principled external controls, artifact removal, and reproducible computation.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.ML
Reinforcement Learning from Human Feedback: A Statistical Perspective
Pangpang Liu, Chengchun Shi and Will Wei Sun
Reinforcement learning from human feedback (RLHF) has emerged as a central framework for aligning large
language models (LLMs) with human preferences. Despite its practical success, RLHF raises fundamental statistical questions because it relies on noisy, subjective, and often heterogeneous feedback to learn reward models and optimize policies. This survey provides a statistical perspective on RLHF, focusing primarily on the LLM
alignment setting. We introduce the main components of RLHF, including supervised fine-tuning, reward modeling, and policy optimization, and relate them to familiar statistical ideas such as Bradley-Terry-Luce (BTL) model, latent utility estimation, active learning, experimental design, and uncertainty quantification. We review methods for learning reward functions from pairwise preference data and for optimizing policies through both two-stage RLHF pipelines and emerging one-stage approaches such as direct preference optimization. We further discuss recent extensions including reinforcement learning from AI feedback, inference-time
algorithms, and reinforcement learning from verifiable rewards, as well as
benchmark datasets, evaluation protocols, and open-source frameworks that support RLHF research. We conclude by highlighting open challenges in RLHF. An accompanying GitHub demo https://github.com/Pangpang-Liu/RLHF_demo illustrates key components of the RLHF pipeline.
Read the full abstract ▾
Machine Learning · Cross-listed from math.OC
Optimal Projection-Free Adaptive SGD for Matrix Optimization
Dmitry Kovalev
Recently, Jiang et al. [2026] developed Leon, a practical variant of One-sided Shampoo [Xie et al., 2025a, An et al., 2025]
algorithm for online convex optimization, which does not require computing a costly quadratic projection at each iteration. Unfortunately, according to the existing analysis, Leon requires tuning an additional hyperparameter in its preconditioner and cannot achieve dimension-independent convergence guarantees for convex optimization problems beyond the bounded gradients assumption. In this paper, we resolve this issue by proving certain stability properties of Leon's preconditioner. Using our improved analysis, we show that tuning the extra hyperparameter can be avoided and, more importantly, develop the first practical variant of One-sided Shampoo with Nesterov acceleration, which does not require computing projections at each iteration. As a side contribution, we obtain improved dimension-independent rates in the non-smooth non-convex setting and develop a unified analysis of the proposed
algorithm, which yields accelerated projection-free adaptive SGD with (block-)diagonal preconditioners.
Read the full abstract ▾
Machine Learning
Causal-Audit: A Framework for Risk Assessment of Assumption Violations in Time-Series Causal Discovery
Marco Ruiz, Miguel Arana-Catania, David R. Ardila et al. (+1 more)
Time-series causal discovery methods rely on assumptions such as stationarity, regular sampling, and bounded temporal dependence. When these assumptions are violated, structure learning can produce confident but misleading causal graphs without warning. We introduce Causal-Audit, a framework that formalizes assumption validation as calibrated risk assessment. The framework computes effect-size diagnostics across five assumption families (stationarity, irregularity, persistence, nonlinearity, and confounding proxies), aggregates them into four calibrated risk scores with uncertainty intervals, and applies an abstention-aware decision policy that recommends methods (e.g., PCMCI+, VAR-based Granger causality) only when evidence supports reliable inference. The semi-automatic diagnostic stage can also be used independently for structured assumption
auditing in individual studies. Evaluation on a synthetic atlas of 500 data-generating processes (DGPs) spanning 10 violation families demonstrates well-calibrated risk scores (AUROC > 0.95), a 62% false positive reduction among recommended
datasets, and 78% abstention on severe-violation cases. On 21 external evaluations from TimeGraph (18 categories) and CausalTime (3 domains), recommend-or-abstain decisions are consistent with
benchmark specifications in all cases. An open-source implementation of our framework is available.
Read the full abstract ▾
All authors Marco Ruiz, Miguel Arana-Catania, David R. Ardila, Rodrigo Ventura
Machine Learning · Cross-listed from cs.CL
Failing to Falsify: Evaluating and Mitigating Confirmation Bias in Language Models
Ayush Rajesh Jhaveri, Anthony GX-Chen, Ilia Sucholutsky et al. (+1 more)
Confirmation bias, the tendency to seek evidence that supports rather than challenges one's belief, hinders one's
reasoning ability. We examine whether large
language models (LLMs) exhibit confirmation bias by adapting the rule-discovery study from human psychology: given a sequence of three numbers (a "triple"), an
agent engages in an interactive feedback loop where it (1) proposes a new triple, (2) receives feedback on whether it satisfies the hidden rule, and (3) guesses the rule. Across eleven LLMs of multiple families and scales, we find that LLMs exhibit confirmation bias, often proposing triples to confirm their hypothesis rather than trying to falsify it. This leads to a slower and less frequent discovery of the hidden rule. We further explore intervention strategies developed for humans (e.g., encouraging the
agent to consider counterexamples). We find that prompting LLMs with such instructions consistently decreases confirmation bias in LLMs, improving rule discovery rates from 42% to 56% on average. Lastly, we mitigate confirmation bias by distilling intervention-induced behavior into LLMs, showing generalization to a new domain (the Blicket test) and beyond rule discovery (mathematical self-correction). Our work shows that confirmation bias is a limitation of LLMs in hypothesis exploration, and that it can be mitigated via injecting interventions designed for humans.
Read the full abstract ▾
All authors Ayush Rajesh Jhaveri, Anthony GX-Chen, Ilia Sucholutsky, Eunsol Choi
Machine Learning
SEDGE: Structural Extrapolated Data Generation
Kun Zhang, Jiaqi Sun, Yiqing Li et al. (+3 more)
This paper aims to address the challenge of data generation beyond the training data and proposes a framework for Structural Extrapolated Data GEneration (SEDGE) based on suitable assumptions on the underlying data-generating process. We provide conditions under which data satisfying novel specifications can be generated reliably, together with the approximate identifiability of the distribution of such data under certain “conservative" assumptions, as well as the inherent non-identifiability of this distribution without such assumptions. On the algorithmic side, we develop practical methods to achieve extrapolated data generation, based on a structure-informed optimization strategy or
diffusion posterior sampling, respectively. We verify the extrapolation performance on synthetic data and also consider extrapolated image generation as a real-world scenario to illustrate the validity of the proposed framework.
Read the full abstract ▾
All authors Kun Zhang, Jiaqi Sun, Yiqing Li, Ignavier Ng, Namrata Deka, Shaoan Xie
Machine Learning · Cross-listed from cs.CV
Guideline2Graph: Profile-Aware Multimodal Parsing for Executable Clinical Decision Graphs
Onur Selim Kilic, Yeti Z. Gurbuz, Cem O. Yaldiz et al. (+4 more)
Clinical practice guidelines are long,
multimodal documents whose branching recommendations are difficult to convert into executable clinical decision support (CDS), and one-shot parsing often breaks cross-page continuity. Recent LLM/VLM extractors are mostly local or text-centric, under-specifying section interfaces and failing to consolidate cross-page control flow across full documents into one coherent decision graph. We present a decomposition-first pipeline that converts full-guideline evidence into an executable clinical decision graph through
topology-aware chunking, interface-constrained chunk graph generation, and provenance-preserving global aggregation. Rather than relying on single-pass generation, the pipeline uses explicit entry/terminal interfaces and semantic deduplication to preserve cross-page continuity while keeping the induced control flow auditable and structurally consistent. We evaluate on an adjudicated prostate-guideline
benchmark with matched inputs and the same underlying VLM backbone across compared methods. On the complete merged graph, our approach improves edge and triplet precision/recall from $19.6\%/16.1\%$ in existing models to $69.0\%/87.5\%$, while node recall rises from $78.1\%$ to $93.8\%$. These results support decomposition-first, auditable guideline-to-CDS conversion on this
benchmark, while current evidence remains limited to one adjudicated prostate guideline and motivates broader multi-guideline validation.
Read the full abstract ▾
All authors Onur Selim Kilic, Yeti Z. Gurbuz, Cem O. Yaldiz, Afra Nawar, Etrit Haxholli, Ogul Can, Eli Waxman
Machine Learning
Time-Warping Recurrent Neural Networks for Transfer Learning
Jonathon Hirschi
Dynamical systems describe how a physical system evolves over time. Physical processes can evolve faster or slower in different environmental conditions. We use time-warping as rescaling the time in a model of a physical system. This thesis proposes a new method of transfer learning for Recurrent Neural Networks (RNNs) based on time-warping. We prove that for a class of linear, first-order differential equations known as time lag models, an LSTM can approximate these systems with any desired accuracy, and the model can be time-warped while maintaining the approximation accuracy. The Time-Warping method of transfer learning is then evaluated in an applied problem on predicting fuel moisture content (FMC), an important concept in wildfire modeling. An RNN with LSTM recurrent layers is pretrained on fuels with a characteristic time scale of 10 hours, where there are large quantities of data available for training. The RNN is then modified with transfer learning to generate predictions for fuels with characteristic time scales of 1 hour, 100 hours, and 1000 hours. The Time-Warping method is evaluated against several known methods of transfer learning. The Time-Warping method produces predictions with an accuracy level comparable to the established methods, despite modifying only a small fraction of the parameters that the other methods modify.
Read the full abstract ▾
Machine Learning
VALOR: Value-Aware Revenue Uplift Modeling with Treatment-Gated Representation for B2B Sales
Vamshi Guduguntla, Kavin Soni and Debanshu Das
B2B sales organizations must identify "persuadable" accounts within zero-inflated revenue distributions to optimize expensive human resource allocation. Standard uplift frameworks struggle with treatment signal collapse in high-dimensional spaces and a misalignment between regression calibration and the ranking of high-value "whales." We introduce VALOR (Value Aware Learning of Optimized (B2B) Revenue), a unified framework featuring a Treatment-Gated Sparse-Revenue Network that uses bilinear interaction to prevent causal signal collapse. The framework is optimized via a novel Cost-Sensitive Focal-ZILN objective that combines a focal mechanism for distributional robustness with a value-weighted ranking loss that scales penalties based on financial magnitude. To provide interpretability for high-touch sales programs, we further derive Robust ZILN-GBDT, a tree based variant utilizing a custom splitting criterion for uplift heterogeneity. Extensive evaluations confirm VALOR's dominance, achieving a 20% improvement in rankability over state-of-the-art methods on public
benchmarks and delivering a validated 2.7x increase in incremental revenue per account in a rigorous 4-month production A/B test.
Read the full abstract ▾
Machine Learning
On the Geometric Structure of Layer Updates in Deep Language Models
Jun-Sik Yoo
We study the
geometric structure of layer updates in deep
language models. Rather than analyzing what information is encoded in intermediate representations, we ask how representations change from one layer to the next. We show that layerwise updates admit a decomposition into a dominant tokenwise component and a residual that is not captured by restricted tokenwise function classes. Across multiple architectures, including
Transformers and state-space models, we find that the full layer update is almost perfectly aligned with the tokenwise component, while the residual exhibits substantially weaker
alignment, larger angular deviation, and significantly lower projection onto the dominant tokenwise subspace. This indicates that the residual is not merely a small correction, but a geometrically distinct component of the transformation. This
geometric separation has functional consequences: approximation error under the restricted tokenwise model is strongly associated with output perturbation, with Spearman correlations often exceeding 0.7 and reaching up to 0.95 in larger models. Together, these results suggest that most layerwise updates behave like structured reparameterizations along a dominant direction, while functionally significant computation is concentrated in a geometrically distinct residual component. Our framework provides a simple, architecture-agnostic method for probing the
geometric and functional structure of layer updates in modern
language models.
Read the full abstract ▾
Machine Learning
Do We Need Frontier Models to Verify Mathematical Proofs?
Aaditya Naik, Guruprerana Shabadi, Rajeev Alur et al. (+1 more)
Advances in training, post-training, and inference-time methods have enabled frontier
reasoning models to win gold medals in math competitions and settle challenging open problems. Gaining trust in the responses of these models requires that natural
language proofs be checked for errors. LLM judges are increasingly being adopted to meet the growing demand for evaluating such proofs. While verification is considered easier than generation, what model capability does reliable verification actually require? We systematically evaluate four open-source and two frontier LLMs on
datasets of human-graded natural
language proofs of competition-level problems. We consider two key metrics: verifier accuracy and self-consistency (the rate of agreement across repeated judgments on the same proof). We observe that smaller open-source models are only up to ~10% behind frontier models in accuracy but they are up to ~25% more inconsistent. Furthermore, we see that verifier accuracy is sensitive to
prompt choice across all models. We then demonstrate that the smaller models, in fact, do possess the mathematical capabilities to verify proofs at the level of frontier models, but they struggle to reliably elicit these capabilities with general judging
prompts. Through an LLM-guided
prompt search, we synthesize an ensemble of specialized
prompts that overcome the specific failure modes of smaller models, boosting their performance by up to 9.1% in accuracy and 15.9% in self-consistency. These gains are realized across models and
datasets, allowing models like Qwen3.5-35B to perform on par with frontier models such as Gemini 3.1 Pro for proof verification.
Read the full abstract ▾
All authors Aaditya Naik, Guruprerana Shabadi, Rajeev Alur, Mayur Naik
Machine Learning · Cross-listed from cs.CV
PlayGen-MoG: Framework for Diverse Multi-Agent Play Generation via Mixture-of-Gaussians Trajectory Prediction
Kevin Song
Multi-
agent trajectory generation in team sports requires models that capture both the diversity of possible plays and realistic spatial coordination between players on plays. Standard generative approaches such as Conditional Variational Autoencoders (CVAE) and
diffusion models struggle with this task, exhibiting posterior collapse or convergence to the
dataset mean. Moreover, most trajectory prediction methods operate in a
forecasting regime that requires multiple frames of observed history, limiting their use for play design where only the initial formation is available. We present PlayGen-MoG, an extensible framework for formation-conditioned play generation that addresses these challenges through three design choices: 1/ a Mixture-of-Gaussians (MoG) output head with shared mixture weights across all
agents, where a single set of weights selects a play scenario that couples all players' trajectories, 2/ relative spatial
attention that encodes pairwise player positions and distances as learned
attention biases, and 3/ non-autoregressive prediction of absolute displacements from the initial formation, eliminating cumulative error drift and removing the dependence on observed trajectory history, enabling realistic play generation from a single static formation alone. On American football tracking data, PlayGen-MoG achieves 1.68 yard ADE and 3.98 yard FDE while maintaining full utilization of all 8 mixture components with
entropy of 2.06 out of 2.08, and qualitatively confirming diverse generation without mode collapse.
Read the full abstract ▾
Machine Learning
Matrix Profile for Time-Series Anomaly Detection: A Reproducible Open-Source Benchmark on TSB-AD
Chin-Chia Michael Yeh
Matrix Profile (MP) methods are an interpretable and scalable family of distance-based methods for time-series anomaly detection, but strong
benchmark performance still depends on design choices beyond a vanilla nearest-neighbor profile. This technical report documents an open-source
Matrix Profile for Anomaly Detection (MMPAD) submission to TSB-AD, a
benchmark that covers both univariate and multivariate time series. The submitted system combines pre-sorted multidimensional aggregation, efficient exclusion-zone-aware k-nearest-neighbor (kNN)
retrieval for repeated anomalies, and moving-average post-processing. To serve as a reproducible reference for MP-based anomaly detection on TSB-AD, we detail the released implementation, the hyperparameter settings for the univariate and multivariate tracks, and the corresponding
benchmark results. We further analyze how the system performs on the aggregate leaderboard and across specific
dataset characteristics.The open-source implementation is available at https://github.com/mcyeh/mmpad_tsb.
Read the full abstract ▾
Machine Learning
Mitigating Data Scarcity in Spaceflight Applications for Offline Reinforcement Learning Using Physics-Informed Deep Generative Models
Alex E. Ballentine, Nachiket U. Bapat and Raghvendra V. Cowlagi
The deployment of reinforcement learning (RL)-based controllers on physical systems is often limited by poor generalization to real-world scenarios, known as the simulation-to-reality (sim-to-real) gap. This gap is particularly challenging in spaceflight, where real-world training data are scarce due to high cost and limited planetary exploration data. Traditional approaches, such as system identification and synthetic data generation, depend on sufficient data and often fail due to modeling assumptions or lack of physics-based constraints. We propose addressing this data scarcity by introducing physics-based learning bias in a generative model. Specifically, we develop the Mutual Information-based Split Variational Autoencoder (MI-VAE), a physics-informed VAE that learns differences between observed system trajectories and those predicted by physics-based models. The latent space of the MI-VAE enables generation of synthetic
datasets that respect physical constraints. We evaluate MI-VAE on a planetary lander problem, focusing on limited real-world data and offline RL training. Results show that augmenting
datasets with MI-VAE samples significantly improves downstream RL performance, outperforming standard VAEs in statistical fidelity, sample diversity, and policy success rate. This work demonstrates a scalable strategy for enhancing autonomous controller robustness in complex, data-constrained environments.
Read the full abstract ▾
Machine Learning
Self-Directed Task Identification
Timothy Gould and Sidike Paheding
In this work, we present a novel machine learning framework called Self-Directed Task Identification (SDTI), which enables models to autonomously identify the correct target variable for each
dataset in a zero-shot setting without pre-training. SDTI is a minimal, interpretable framework demonstrating the feasibility of repurposing core machine learning concepts for a novel task structure. To our knowledge, no existing architectures have demonstrated this ability. Traditional approaches lack this capability, leaving data annotation as a time-consuming process that relies heavily on human effort. Using only standard neural network components, we show that SDTI can be achieved through appropriate problem formulation and architectural design. We evaluate the proposed framework on a range of
benchmark tasks and demonstrate its effectiveness in reliably identifying the ground truth out of a set of potential target variables. SDTI outperformed baseline architectures by 14% in F1 score on synthetic task identification
benchmarks. These proof-of-concept experiments highlight the future potential of SDTI to reduce dependence on manual annotation and to enhance the scalability of autonomous learning systems in real-world applications.
Read the full abstract ▾
Machine Learning · Cross-listed from cs.ET
Photonic convolutional neural network with pre-trained in situ training
Saurabh Ranjan, Sonika Thakral and Amit Sehgal
Convolutional neural networks (CNNs) have transformed image processing, but the energy consumption and inference latency of electronic based implementations remain fundamental bottlenecks. These limitations have motivated the search for alternative hardware architectures beyond Complementary metal-
oxide-
semiconductor (CMOS) chips. Optical systems can perform linear
matrix operations at the speed of light with extremely low energy dissipation, making them attractive for CNN acceleration. However, building a fully coherent photonic CNN that performs both linear and nonlinear operations and training it efficiently remains an open challenge. Here we present a fully photonic convolutional neural network (PCNN) that executes image classification in the optical domain, including convolution, max-pooling, nonlinear activation, and fully connected layers. The network achieves 94.49 percent accuracy on the MNIST
dataset distributed across Mach Zehnder Interferometer (MZI) meshes, weighted Multimode Interferometer (MMI) trees, and a microring resonator based nonlinearity. A mathematically exact differentiable digital twin, enables backpropagation for ex situ pre training, reaches 97.45 percent digital accuracy. Trained phases are transferred one-to-one to the photonic hardware and refined via a gradient free
algorithm that estimates the full gradient with only two forward passes. The architecture exhibits inherent robustness to non idealities, under the compound effect of propagation loss, MZI insertion loss, fabrication disorder, and thermal crosstalk. A bottom-up power analysis yields 10.83 W static chip consumption and 843 ns inference latency, translating to 220 to 330 times greater energy efficiency than state of the art electronic GPUs for single-image inference.
Read the full abstract ▾
Machine Learning
Plateaus, Optima, and Overfitting in Multi-Layer Perceptrons: A Saddle-Saddle-Attractor Scenario
Alex Alì Maleknia and Yuzuru Sato
Vanishing gradients and overfitting are central problems in machine learning, yet are typically analyzed in asymptotic regimes that obscure their dynamical origins. Here we provide a dynamical description of learning in multi-layer perceptrons (MLPs) via a minimal model inspired by Fukumizu and Amari. We show that training dynamics traverse plateau and near-optimal regions, both organized by saddle structures, before converging to an overfitting regime. Under suitable conditions on the data, this regime collapses to a single attractor modulo
symmetry. Furthermore, for finite noisy
datasets, convergence to the theoretical optimum is impossible, and the dynamics necessarily settle into an overfitting solution.
Read the full abstract ▾
Machine Learning · Cross-listed from cs.CV
ActionParty: Multi-Subject Action Binding in Generative Video Games
Alexander Pondaven, Ziyi Wu, Igor Gilitschenski et al. (+4 more)
Recent advances in
video diffusion have enabled the development of "world models" capable of simulating interactive environments. However, these models are largely restricted to single-
agent settings, failing to control multiple
agents simultaneously in a scene. In this work, we tackle a fundamental issue of action binding in existing
video diffusion models, which struggle to associate specific actions with their corresponding subjects. For this purpose, we propose ActionParty, an action controllable multi-subject world model for generative
video games. It introduces subject state
tokens, i.e. latent variables that persistently capture the state of each subject in the scene. By jointly modeling state
tokens and
video latents with a spatial biasing mechanism, we disentangle global
video frame rendering from individual action-controlled subject updates. We evaluate ActionParty on the Melting Pot
benchmark, demonstrating the first
video world model capable of controlling up to seven players simultaneously across 46 diverse environments. Our results show significant improvements in action-following accuracy and identity consistency, while enabling robust autoregressive tracking of subjects through complex interactions.
Read the full abstract ▾
All authors Alexander Pondaven, Ziyi Wu, Igor Gilitschenski, Philip Torr, Sergey Tulyakov, Fabio Pizzati, Aliaksandr Siarohin
Machine Learning · Cross-listed from cs.CL
Grounded Token Initialization for New Vocabulary in LMs for Generative Recommendation
Daiwei Chen, Zhoutong Fu, Chengming Jiang et al. (+12 more)
Language models (LMs) are increasingly extended with new learnable vocabulary
tokens for domain-specific tasks, such as Semantic-ID
tokens in generative recommendation. The standard practice initializes these new
tokens as the mean of existing vocabulary
embeddings, then relies on supervised fine-tuning to learn their representations. We present a systematic analysis of this strategy: through spectral and
geometric diagnostics, we show that mean initialization collapses all new
tokens into a degenerate subspace, erasing inter-
token distinctions that subsequent fine-tuning struggles to fully recover. These findings suggest that
token initialization is a key bottleneck when extending LMs with new vocabularies. Motivated by this diagnosis, we propose the Grounded
Token Initialization Hypothesis: linguistically grounding novel
tokens in the pretrained
embedding space before fine-tuning better enables the model to leverage its general-purpose knowledge for novel-
token domains. We operationalize this hypothesis as GTI (Grounded
Token Initialization), a lightweight grounding stage that, prior to fine-tuning, maps new
tokens to distinct, semantically meaningful locations in the pretrained
embedding space using only paired linguistic supervision. Despite its simplicity, GTI outperforms both mean initialization and existing auxiliary-task adaptation methods in the majority of evaluation settings across multiple generative recommendation
benchmarks, including industry-scale and public
datasets. Further analyses show that grounded
embeddings produce richer inter-
token structure that persists through fine-tuning, corroborating the hypothesis that initialization quality is a key bottleneck in vocabulary extension.
Read the full abstract ▾
All authors Daiwei Chen, Zhoutong Fu, Chengming Jiang, Haichao Zhang, Ran Zhou, Tan Wang, Chunnan Yao, Guoyao Li, Rui Cai, Yihan Cao, Ruijie Jiang, Fedor Borisyuk, Jianqiang Shen, Jingwei Wu, Ramya Korlakai Vinayak
Machine Learning
Batched Contextual Reinforcement: A Task-Scaling Law for Efficient Reasoning
Bangji Yang, Hongbo Ma, Jiajun Fan et al. (+1 more)
Large
Language Models employing Chain-of-Thought
reasoning achieve strong performance but suffer from excessive
token consumption that inflates inference costs. Existing efficiency methods such as explicit length penalties, difficulty
estimators, or multi-stage curricula either degrade
reasoning quality or require complex training pipelines. We introduce Batched Contextual Reinforcement, a minimalist, single-stage training paradigm that unlocks efficient
reasoning through a simple structural modification: training the model to solve N problems simultaneously within a shared context window, rewarded purely by per-instance accuracy. This formulation creates an implicit
token budget that yields several key findings: (1) We identify a novel task-scaling law: as the number of concurrent problems N increases during inference, per-problem
token usage decreases monotonically while accuracy degrades far more gracefully than baselines, establishing N as a controllable throughput dimension. (2) BCR challenges the traditional accuracy-efficiency trade-off by demonstrating a "free lunch" phenomenon at standard single-problem inference. Across both 1.5B and 4B model families, BCR reduces
token usage by 15.8% to 62.6% while consistently maintaining or improving accuracy across five major mathematical
benchmarks. (3) Qualitative analyses reveal emergent self-regulated efficiency, where models autonomously eliminate redundant metacognitive loops without explicit length supervision. (4) Crucially, we empirically demonstrate that implicit budget constraints successfully circumvent the
adversarial gradients and catastrophic optimization collapse inherent to explicit length penalties, offering a highly stable, constraint-based alternative for length control. These results prove BCR practical, showing simple structural incentives unlock latent high-density
reasoning in LLMs.
Read the full abstract ▾
All authors Bangji Yang, Hongbo Ma, Jiajun Fan, Ge Liu
Machine Learning · Cross-listed from hep-th
Topological Effects in Neural Network Field Theory
Christian Ferko, James Halverson, Vishnu Jejjala et al. (+1 more)
Neural network field theory formulates field theory as a statistical ensemble of fields defined by a network architecture and a density on its parameters. We extend the construction to topological settings via the inclusion of discrete parameters that label the topological
quantum number. We recover the Berezinskii–Kosterlitz–Thouless transition, including the
spin-wave critical line and the proliferation of
vortices at high temperatures. We also verify the T-duality of the bosonic string, showing invariance under the exchange of momentum and winding on $S^1$, the transformation of the sigma model couplings according to the Buscher rules on constant toroidal backgrounds, the enhancement of the current
algebra at self-dual radius, and non-
geometric T-fold transition functions.
Read the full abstract ▾
All authors Christian Ferko, James Halverson, Vishnu Jejjala, Brandon Robinson
Machine Learning
go-$m$HC: Direct Parameterization of Manifold-Constrained Hyper-Connections via Generalized Orthostochastic Matrices
Torque Dandachi and Sophia Diggs-Galligan
Doubly
stochastic matrices enable learned mixing across residual streams, but parameterizing the set of doubly
stochastic matrices (the Birkhoff polytope) exactly and efficiently remains an open challenge. Existing exact methods scale factorially with the number of streams ($d$), while Kronecker-factorized approaches are efficient but expressivity-limited. We introduce a novel exact parameterization grounded in the theory of generalized orthostochastic
matrices, which scales as $\mathcal{O}(d^3)$ and exposes a single hyperparameter $s$ which continuously interpolates between a computationally efficient boundary and the fully expressive Birkhoff polytope. Building on
Manifold-Constrained Hyper-Connections ($m$HC), a framework for learned dynamic layer connectivity, we instantiate this parameterization in go-$m$HC. Our method composes naturally with Kronecker-factorized methods, substantially recovering expressivity at similar FLOP costs. Spectral analysis indicates that go-$m$HC fills the Birkhoff polytope far more completely than Kronecker-factorized baselines. On synthetic stream-mixing tasks, go-$m$HC achieves the minimum theoretical loss while converging up to $10\times$ faster. We validate our approach in a 30M parameter GPT-style
language model. The expressivity, efficiency, and exactness of go-$m$HC offer a practical avenue for scaling $d$ as a new dimension of model capacity.
Read the full abstract ▾
Machine Learning
Taming the Exponential: A Fast Softmax Surrogate for Integer-Native Edge Inference
Dimitrios Danopoulos, Enrico Lupi, Michael Kagan et al. (+1 more)
Softmax can become a computational bottleneck in the
Transformer model's Multi-Head
Attention (MHA) block, particularly in small models under low-precision inference, where exponentiation and normalization incur significant overhead. As such, we suggest using Head-Calibrated Clipped-Linear Softmax (HCCS), a bounded, monotone surrogate to the exponential softmax function, which uses a clipped linear mapping of the max centered
attention logits. This approximation produces a stable probability distribution, maintains the ordering of the original logits and has non-negative values. HCCS differs from previous softmax surrogates as it includes a set of lightweight calibration parameters that are optimized offline based on a representative
dataset and calibrated for each individual
attention head to preserve the statistical properties of the individual heads. We describe a hardware-motivated implementation of HCCS for high-throughput scenarios targeting the AMD Versal AI Engines. The current reference implementations from AMD for this platform rely upon either bfloat16 arithmetic or LUTs to perform the exponential operation, which might limit the throughput of the platform and fail to utilize the high-throughput integer vector processing units of the AI Engine. In contrast, HCCS provides a natural mapping to the AI Engines' int8 multiply accumulate (MAC) units. To the best of our knowledge, this is the first int8 optimized softmax surrogate for AMD AI engines that significantly exceeds the speed performance of other reference implementations while maintaining competitive task accuracy on small or heavily quantized MHA workloads after quantization-aware retraining.
Read the full abstract ▾
All authors Dimitrios Danopoulos, Enrico Lupi, Michael Kagan, Maurizio Pierini
Machine Learning
Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing
Gengsheng Li, Tianyu Yang, Junfeng Fang et al. (+6 more)
Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large
language models. While Group Relative Policy Optimization (GRPO) is widely adopted, its coarse credit assignment uniformly penalizes failed rollouts, lacking the
token-level focus needed to efficiently address specific deviations. Self-
Distillation Policy Optimization (SDPO) addresses this by providing denser, more targeted logit-level supervision that facilitates rapid early improvement, yet it frequently collapses during prolonged training. We trace this late-stage instability to two intrinsic flaws: self-
distillation on already-correct samples introduces optimization ambiguity, and the self-teacher's signal reliability progressively degrades. To resolve these issues, we propose Sample-Routed Policy Optimization (SRPO), a unified on-policy framework that routes correct samples to GRPO's reward-aligned reinforcement and failed samples to SDPO's targeted logit-level correction. SRPO further incorporates an
entropy-aware dynamic weighting mechanism to suppress high-
entropy, unreliable
distillation targets while emphasizing confident ones. Evaluated across five
benchmarks and two model scales, SRPO achieves both the rapid early improvement of SDPO and the long-horizon stability of GRPO. It consistently surpasses the peak performance of both baselines, raising the five-
benchmark average on Qwen3-8B by 3.4% over GRPO and 6.3% over SDPO, while simultaneously yielding moderate response lengths and lowering per-step compute cost by up to 17.2%.
Read the full abstract ▾
All authors Gengsheng Li, Tianyu Yang, Junfeng Fang, Mingyang Song, Mao Zheng, Haiyun Guo, Dan Zhang, Jinqiao Wang, Tat-Seng Chua
Machine Learning · Cross-listed from cs.AI
De Jure: Iterative LLM Self-Refinement for Structured Extraction of Regulatory Rules
Keerat Guliani, Deepkamal Gill, David Landsman et al. (+3 more)
Regulatory documents encode legally binding obligations that LLM-based systems must respect. Yet converting dense, hierarchically structured legal text into machine-readable rules remains a costly, expert-intensive process. We present De Jure, a fully automated, domain-agnostic pipeline for extracting structured regulatory rules from raw documents, requiring no human annotation, domain-specific prompting, or annotated gold data. De Jure operates through four sequential stages: normalization of source documents into structured Markdown; LLM-driven semantic decomposition into structured rule units; multi-criteria LLM-as-a-judge evaluation across 19 dimensions spanning metadata, definitions, and rule semantics; and iterative repair of low-scoring extractions within a bounded regeneration budget, where upstream components are repaired before rule units are evaluated. We evaluate De Jure across four models on three regulatory corpora spanning finance, healthcare, and AI governance. On the finance domain, De Jure yields consistent and monotonic improvement in extraction quality, reaching peak performance within three judge-guided iterations. De Jure generalizes effectively to healthcare and AI governance, maintaining high performance across both open- and closed-source models. In a downstream compliance question-answering evaluation via RAG, responses grounded in De Jure extracted rules are preferred over prior work in 73.8% of cases at single-rule
retrieval depth, rising to 84.0% under broader
retrieval, confirming that extraction fidelity translates directly into downstream utility. These results demonstrate that explicit, interpretable evaluation criteria can substitute for human annotation in complex regulatory domains, offering a scalable and auditable path toward regulation-grounded LLM
alignment.
Read the full abstract ▾
All authors Keerat Guliani, Deepkamal Gill, David Landsman, Nima Eshraghi, Krishna Kumar, Lovedeep Gondara
Machine Learning
Crystalite: A Lightweight Transformer for Efficient Crystal Modeling
Tin Hadži Veljković, Joshua Rosenthal, Ivor Lončarić et al. (+1 more)
Generative models for crystalline materials often rely on equivariant graph neural networks, which capture
geometric structure well but are costly to train and slow to sample. We present Crystalite, a lightweight
diffusion Transformer for
crystal modeling built around two simple inductive biases. The first is Subatomic Tokenization, a compact chemically structured atom representation that replaces high-dimensional one-hot encodings and is better suited to continuous
diffusion. The second is the Geometry Enhancement Module (GEM), which injects periodic minimum-image pair geometry directly into
attention through additive
geometric biases. Together, these components preserve the simplicity and efficiency of a standard
Transformer while making it better matched to the structure of crystalline materials. Crystalite achieves state-of-the-art results on
crystal structure prediction
benchmarks, and de novo generation performance, attaining the best S.U.N. discovery score among the evaluated baselines while sampling substantially faster than geometry-heavy alternatives.
Read the full abstract ▾
All authors Tin Hadži Veljković, Joshua Rosenthal, Ivor Lončarić, Jan-Willem van de Meent
Machine Learning
SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization
Zhengxi Lu, Zhiyuan Yao, Jinyang Wu et al. (+7 more)
Agent skills, structured packages of procedural knowledge and executable resources that
agents dynamically load at inference time, have become a reliable mechanism for augmenting LLM
agents. Yet inference-time skill augmentation is fundamentally limited:
retrieval noise introduces irrelevant guidance, injected skill content imposes substantial
token overhead, and the model never truly acquires the knowledge it merely follows. We ask whether skills can instead be internalized into model parameters, enabling zero-shot autonomous behavior without any runtime skill
retrieval. We introduce SKILL0, an in-context reinforcement learning framework designed for skill internalization. SKILL0 introduces a training-time curriculum that begins with full skill context and progressively withdraws it. Skills are grouped offline by category and rendered with interaction history into a compact visual context, teaching he model tool invocation and multi-turn task completion. A Dynamic Curriculum then evaluates each skill file's on-policy helpfulness, retaining only those from which the current policy still benefits within a linearly decaying budget, until the
agent operates in a fully zero-shot setting. Extensive agentic experiments demonstrate that SKILL0 achieves substantial improvements over the standard RL baseline (+9.7% for ALFWorld, +6.6% for Search-QA, and+10.1% for WebShop), while maintaining a highly efficient context of fewer than 0.5k
tokens per step. Our code is available at https://github.com/ZJU-REAL/SkillZero.
Read the full abstract ▾
All authors Zhengxi Lu, Zhiyuan Yao, Jinyang Wu, Chengcheng Han, Qi Gu, Xunliang Cai, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen
Machine Learning
Model-Based Reinforcement Learning for Control under Time-Varying Dynamics
Klemens Iten, Bruce Lee, Chenhao Li et al. (+3 more)
Learning-based control methods typically assume stationary system dynamics, an assumption often violated in real-world systems due to drift, wear, or changing operating conditions. We study reinforcement learning for control under time-varying dynamics. We consider a continual model-based reinforcement learning setting in which an
agent repeatedly learns and controls a dynamical system whose transition dynamics evolve across episodes. We analyze the problem using Gaussian process dynamics models under frequentist variation-budget assumptions. Our analysis shows that persistent non-stationarity requires explicitly limiting the influence of outdated data to maintain calibrated uncertainty and meaningful dynamic regret guarantees. Motivated by these insights, we propose a practical optimistic model-based reinforcement learning
algorithm with adaptive data buffer mechanisms and demonstrate improved performance on continuous control
benchmarks with non-stationary dynamics.
Read the full abstract ▾
All authors Klemens Iten, Bruce Lee, Chenhao Li, Lenart Treven, Andreas Krause, Bhavya Sukhija