A gray-box workflow, PC-MCMC-CIGP, integrates spike-and-slab topology sampling, hard conservation and thermodynamic screening, and a Chemical-Informed Gaussian Process (CIGP) residual model for extracting interpretable governing equations from sparse noisy chemical time-series data. On the H2+Br2 benchmark, it distinguishes elementary radical pathways; on styrene epoxidation, it improves yield by 12.5% over the baseline. A 10-seed acquisition study reveals trade-offs: PC-EI reduces low-yield BO suggestions, while EI-style criteria yield the strongest final performance.

A new approach to physical neural networks places trainable nonlinear functions on connections rather than using scalar weights, achieving low-power continuous control tasks with far fewer nodes. The design, implemented on analogue arrays, shows task-dependent benefits and projects to 30 microwatts in CMOS.

This paper presents a comprehensive survey on federated causal discovery and inference, proposing multi-dimensional taxonomies and highlighting the integration of causal structure learning and effect estimation in a unified pipeline, addressing challenges like privacy and data heterogeneity.

This paper investigates whether six offline RL losses (SFT, RFT, DFT, RIFT, Offline GRPO, DPO) are mechanistically distinct in weight-space geometry when used for reasoning distillation. Using identical math rollouts from Qwen3-4B, they find SFT, RFT, and RIFT have nearly colinear deltas; DFT diverges; Offline GRPO adds orthogonal components; and DPO lies in a near-orthogonal subspace with highest accuracy but a mode-connectivity barrier.

This paper presents an automated large-scale search pipeline for heterogeneous 4-Expert Mixture-of-Experts (MoE4) architectures within the LEMUR neural network dataset ecosystem. Over 28 days on an NVIDIA RTX 4090, the pipeline generated 4,463 candidate models and evaluated 1,021. A critical coverage bias was discovered: due to alphabetical enumeration, the search space was anchored to the AirNet family. Within this scope, ShuffleNet and MobileNetV3 ensembles achieved highest average accuracy of 0.632, while FractalNet and MNASNet were identified as low-yield families.

This study constructs large-scale algorithm co-occurrence networks in NLP using deep learning on full-text papers. It analyzes network structure and centrality to assess collective influence over decades, finding that classic, high-performing, and cross-period algorithms dominate, and declining influence first loses core network position.

A new method called SGPO improves LLM reasoning by replacing trajectory imitation with strategy distillation, achieving better generalization and outperforming baselines on math benchmarks.

A hybrid model using ensemble feature selection (ANOVA and mutual information) and Harris Hawks optimization-tuned logistic regression predicts mental health risk in female sex workers (FSWs). Achieved 95.78% accuracy on 3,005 FSWs, identifying post-traumatic stress, client violence, and occupational factors as key depression drivers. XAI enables early intervention and targeted care.

Recommender systems often induce filter bubbles and semantic homogenization by optimizing solely for immediate engagement. This paper introduces a multi-objective reinforcement learning framework that treats engagement, diversity, and fairness as distinct reward signals using a Pareto-DQN agent. Experiments on MovieLens show that hypervolume-based action selection disrupts feedback loops leading to semantic collapse, achieving societal gains with minimal engagement impact.

This paper investigates whether language model (LM) agents can assist in explaining circuit components after they have been localized in mechanistic interpretability. The authors introduce AgenticInterpBench, a benchmark of 84 semi-synthetic transformer circuits with 163 component-level annotations, and HyVE (Hypothesize, Validate, Explain), an agentic explainer that iteratively observes, hypothesizes, and causally validates. Experiments across four LM backbones show that HyVE recovers useful explanations, but no backbone is uniformly best; failures mainly occur in the validation step. A case study on an arithmetic circuit in Llama-3-8B demonstrates extension to naturally trained models. LM agents are promising but reliable validation remains a key obstacle.

A new study shows that reinforcement learning on beneficial behavior in realistic domains can produce broad and persistent alignment generalization, with interventions limited to health improving non-health alignment evaluations and resistance to adversarial attacks.

Researchers propose a hierarchical multi-agent reinforcement learning framework that enforces hard safety constraints via a constraint manifold at low level while enabling effective coordination through high-level policy learning. The approach provides theoretical safety guarantees, stationary learning dynamics, and achieves competitive performance with nearly perfect safety rates and strong generalization.

This paper explores the nature of AI agents, distinguishing between 'agentic' systems with engineered workflows and 'agentive' systems with endogenous capabilities. It proposes the Goal-Identity-Configurator (GIC) architecture and emphasizes auditability, controllability, and safety of autonomous systems under human oversight.

A new framework extracts rule-grounded reasoning traces from classical planners to supervise driving VLA models, ensuring structurally coupled reasoning and motion generation, with significant performance gains.

RIFT-Bench is a graph representation-driven methodology for dynamic red-teaming that enables unified security evaluations across diverse agentic AI architectures. It operates in two automated phases—Discovery and Scanning—and supports evaluation of mitigation strategies, demonstrating effectiveness across 45 systems.

NVIDIA and AWS collaborate to provide scalable, low-latency AI infrastructure with new EC2 G7 instances featuring Blackwell GPUs, GPU-accelerated vector indexing in OpenSearch Serverless powered by cuVS, and AWS achieving NVIDIA Exemplar Cloud status for GB300 training.

Scaling laws are one of the most critical empirical findings in deep learning, describing power-law relationships between model size, data, compute, and loss. This article reviews the development from early theory to modern empirical studies, including Kaplan et al.'s classic scaling laws and the Chinchilla scaling laws, and discusses key findings such as compute-optimal allocation.