Research→Official→arXiv Machine Learning
A new preprint introduces a white-box instrument based on hidden deterministic finite automata to separately measure reward attainment and latent-state learning in reinforcement learning (RL) agents. The study demonstrates that high reward does not necessarily indicate that an agent has learned the underlying task state, distinguishing between 'perception gaps' (where latent state is not recoverable from observations) and 'planning gaps' (where state is recoverable but not used). The authors show that optimizer strength, task structure, and observation informativeness each influence the relationship between reward and state learning.
Why it matters: This work challenges the common assumption that high reward in RL implies genuine task understanding, providing a method to diagnose when agents exploit shortcuts rather than learning true latent states.
Jul 15, 2026
Research→Official→arXiv Machine Learning
CARE-LoRA is a method designed to reduce the memory bottleneck of activations during LoRA fine-tuning by replacing full input activations with low-rank compressed activations from the LoRA branch. It introduces a lightweight reconstruction matrix computed during the forward pass, enabling gradient reconstruction in backpropagation while keeping LoRA matrices fully trainable. Experiments across various models and tasks show that CARE-LoRA achieves competitive or superior performance compared to standard LoRA and its variants, with a substantially reduced memory footprint.
Why it matters: This approach addresses a major memory limitation in fine-tuning large models, potentially making such processes more feasible on hardware with limited memory.
Jul 15, 2026
Research→Official→arXiv Machine Learning
Researchers present Support Vector Attention (SV-Attention), a novel max-margin memory mechanism for attention models that enables certified token eviction and exact unlearning. SV-Attention achieves higher rare-item recall (0.86 vs. 0.32) and better retention (0.80 vs. 0.05 deterioration hours) on MIMIC-IV data streams compared to baselines, and demonstrates improved compression (2.178 BPC vs. 2.383) on enwik8 relative to a sliding-window Transformer. The method also supports surgical forgetting, exact editing, and patient-record deletion in real-world scenarios.
Why it matters: This work introduces a principled and practical approach to certified unlearning in attention mechanisms, addressing a key challenge in machine learning privacy and data management.
Jul 15, 2026
Research→Official→arXiv Machine Learning
A new training-free attribution method is introduced to analyze the dependencies of feedforward network (FFN) neurons in Transformer models. The study finds that small, sparse subsets of upstream activations and attention outputs can preserve neuron activations with high fidelity, even when other inputs are masked. This reveals that FFNs, despite their dense parameterization, have sparse and structured inter-layer dependencies at the neuron level. The method is scalable and can be used for circuit-level interpretability and identifying sparse pathways for potential efficiency gains.
Why it matters: This work offers a practical and scalable approach to understanding and potentially optimizing the internal structure of Transformer models, which could lead to more efficient inference and improved interpretability.
Jul 15, 2026
Research→Official→arXiv Machine Learning
Researchers introduce the first pseudo-polynomial-time exact algorithm for computing Data Shapley values in weighted k-nearest-neighbor (KNN) regression, addressing a longstanding computational barrier. The work also presents a certified fully polynomial-time approximation scheme (FPTAS) with machine-checkable error bounds and extends the approach to soft-label multi-class prediction. An open-source implementation and the first exact ground-truth dataset for weighted-regression Data Shapley are provided.
Why it matters: This advance enables deterministic, certified data valuation in weighted KNN regression, providing a reliable reference for auditing and improving approximate methods.
Jul 15, 2026
Research→Official→arXiv Machine Learning
A new preprint introduces an agentic AI scientific community composed of virtual laboratories that autonomously discover neural operator architectures. Each lab uses LLM agents for planning, training, and peer review, operating under a citation-based economy. The system demonstrates the ability to find high-accuracy, low-parameter-count neural operator designs across five PDE benchmark problems, outperforming rule-based alternatives in maintaining architectural diversity.
Why it matters: This work represents a significant advance in automating scientific discovery using LLM-driven agentic systems, highlighting the potential for AI to autonomously innovate in complex domains.
Jul 15, 2026
Research→Official→arXiv Machine Learning
Airbnb has developed and deployed Proximity Features, a privacy-compliant system that addresses the cold-start personalization problem by grouping users based on geographic proximity using geo-IP data and adaptive clustering. This approach aggregates signals for groups of around 1,000 users, avoiding the need for persistent individual identifiers and operating within consent-gated privacy controls. Online A/B tests in production show statistically significant increases in bookings, particularly for users lacking recent or any history.
Why it matters: This work provides a scalable, privacy-preserving solution to cold-start personalization with demonstrated real-world impact at a major online platform.
Jul 15, 2026
Research→Official→arXiv Machine Learning
Researchers introduce LiteTopK, a fused GPU kernel designed to accelerate Indexer-TopK operations in long-context sparse attention by leveraging the curse of dimensionality. LiteTopK samples data to estimate score ranges and partitions candidates into bins, reducing memory traffic and overhead while maintaining exact top-k correctness. Experiments show a 1.2x speedup in the prefill stage of GLM 5.2 during real-world deployment, along with lower memory usage.
Why it matters: This work offers a practical advance in the efficiency of sparse attention for large language models, enabling faster and more memory-efficient processing of long contexts.
Jul 15, 2026
Research→Official→arXiv Machine Learning
A new preprint demonstrates that up to 77% of features recovered by sparse autoencoders (SAEs) with high cosine similarity to ground-truth directions are causally inert, meaning the matched atom does not activate when the feature is present. The authors introduce sae-causal-audit, a model-agnostic tool for causal validation, and identify two types of inertness: structural (due to antipodal-pair geometry) and competitive (arising from TopK pathologies in degraded SAEs).
Why it matters: This work challenges the use of correlational metrics alone for evaluating SAE interpretability, showing that high recovery scores can mask a lack of causal relevance, which is important for mechanistic interpretability research.
Jul 15, 2026
Research→Official→arXiv Machine Learning
A new audit of six KV-cache compression methods finds that their performance rankings reverse when compression is performed before seeing the query (query-agnostic) compared to after (query-aware). In the more realistic query-agnostic setting, only KeyDiff consistently outperforms trivial baselines, while the widely used SnapKV method underperforms a simple 'keep start and recent window' baseline.
Why it matters: This study challenges the validity of current evaluation protocols for KV-cache compression, suggesting that some popular methods may be less effective in real-world LLM deployments than previously thought.
Jul 15, 2026
Research→Official→arXiv Information Retrieval
A new study demonstrates that applying Principal Component Analysis (PCA) to compress query embeddings can enhance retrieval performance in specialized domains. The approach was tested across 9 dense retrievers and 14 MTEB datasets, showing improved NDCG@10 scores in 75.4% of model-dataset pairs compared to standard embeddings. This method preserves domain-relevant features while removing non-discriminative components, providing a lightweight, training-free alternative to traditional domain adaptation methods.
Why it matters: This technique offers a simple and efficient way to improve dense retrieval in specialized domains without the need for costly annotation or retraining.
Jul 15, 2026
Research→Official→arXiv Machine Learning
A new method equips speculative execution in LLM agents with three online memory systems—a contrastive transition table, episodic memory, and confusion tracker—to improve prediction quality by learning from past agent trajectories. The approach achieves 19–39% relative accuracy improvement on action prediction and up to 2.5× increase on observation prediction tasks, with gains increasing as memory accumulates. All speculation occurs during idle time, resulting in no added wall-clock cost and preserving identical agent trajectories compared to non-speculative execution.
Why it matters: This work demonstrates a way for LLM agents to learn from experience and improve efficiency without sacrificing correctness, potentially advancing the deployment of faster and more capable agentic systems.
Jul 15, 2026
Research→Official→arXiv Information Retrieval
Researchers have introduced iTIMO, a dataset designed to address the underexplored task of travel itinerary modification. The dataset is generated by instructing large language models to perturb real-world itineraries using REPLACE, ADD, and DELETE operations, guided by intents such as popularity disruptions, spatial distance, and category diversity. The work also benchmarks state-of-the-art LLMs on this task, revealing their strengths and limitations.
Why it matters: iTIMO provides a comprehensive testbed for developing and evaluating adaptive travel recommendation systems that can handle dynamic and changing user needs.
Jul 15, 2026
Research→Official→arXiv Information Retrieval
Researchers introduce SkillSelect-Serve, a framework that recommends reusable agent skills for LLM agents while respecting constraints such as token budget, risk, and tool availability. By profiling skills as structured services and applying a constrained projection method, the system achieves 100% deliverability with minimal loss in hit rate. The approach outperforms traditional Top-k retrieval methods, significantly reducing risk exposure and tool violations.
Why it matters: This work offers a practical advance for deploying LLM agents by enabling reliable, constraint-aware skill selection, which is crucial for efficient operation under real-world resource limits.
Jul 15, 2026
Research→Official→arXiv Machine Learning
Researchers introduce FastAlign, a sparsity-aware framework for optimal transport-based network alignment. FastAlign maintains alignment quality comparable to state-of-the-art methods while significantly improving computational efficiency, achieving up to 9.45x speedup on CPU and 32.54x on GPU. The method leverages mixed sparse-dense operations and custom kernel fusion to address scalability challenges in large-scale network alignment.
Why it matters: This work enables efficient analysis of large-scale networks, addressing a major scalability bottleneck in network alignment tasks relevant to fields such as social network analysis and fraud detection.
Jul 15, 2026
Research→Official→arXiv Computers and Society
A new preprint finds that large language models (LLMs) generate responses that are more concentrated and mainstream than the diverse, long-tail outputs produced by humans. The researchers tested interventions such as increasing temperature sampling, prompting for diverse perspectives, and aggregating outputs from multiple models, finding that these methods can improve diversity, but single-model outputs still fall short of human-level diversity.
Why it matters: The study raises concerns that LLMs could reduce cultural diversity in generated content, which has implications for AI policy and the preservation of democratic values.
Jul 15, 2026
Policy Safety→Official→arXiv Computers and Society
A new benchmark, NOHARM, evaluates 20 large language models (LLMs) and 4 clinical AI tools on 1,100 medical consultation cases, finding that direct use of AI-generated recommendations could result in severe harm in up to 24.6% of cases, with omission errors accounting for over 80% of severe errors. In a randomized study of 101 physicians, AI assistance improved performance, but physicians often omitted valuable AI recommendations, indicating complementary strengths in human-AI teaming.
Why it matters: This study provides the first systematic measurement of clinical safety in LLM-generated medical advice, revealing that widely used AI tools can produce potentially harmful recommendations and highlighting the need for explicit safety evaluation.
Jul 15, 2026
Research→Official→arXiv Computers and Society
Researchers introduce EG-VAR, a Lean 4-based tool-calling architecture that ensures every verified output is grounded in attested tool calls and kernel-checked inference, thereby eliminating unsupported claims. On a subset of TableBench numerical reasoning tasks (n=120), EG-VAR achieves perfect accuracy (120/120) compared to a 95% baseline, and maintains 100% source-faithfulness on stress tests where baselines drop to 80-90%. The system also provides explicit audit trails for abstentions and formalization errors.
Why it matters: EG-VAR offers a practical and auditable approach to eliminating LLM hallucination in empirical inference, potentially transforming trustworthiness in high-stakes AI applications.
Jul 15, 2026
Research→Official→arXiv Information Retrieval
MESH is a unified retrieval scaling framework designed to address the scaling bias of heterogeneity in large-scale recommendation systems. Through a modularized architecture with gated bias correction, MESH partitions the feature space to reduce interference between different content types, leading to a 14x improvement in the power-law scaling exponent for fresh items. In online evaluations on Pinterest's Related Pins platform, MESH demonstrated a +5.5% lift in fresh-item repins, a 55% improvement in funnel efficiency, and a +0.46% improvement in user retention.
Why it matters: MESH represents a significant advance in unifying fragmented retrieval systems, improving scalability and performance for diverse content in large-scale recommendation platforms.
Jul 15, 2026
Research→Official→arXiv Computers and Society
A new preprint compares large language models (LLMs) such as GPT, Twitter-roBERTa, and LLaMA to traditional machine learning methods for analyzing open-ended survey responses. The study finds that LLMs achieve higher classification accuracy, especially in sentiment and thematic analysis, but exhibit significant variation in consistency and the explicitness of their reasoning. These results highlight important trade-offs between predictive performance and interpretability in large-scale qualitative research.
Why it matters: The study offers practical insights for researchers seeking to balance automation with interpretive rigor when applying LLMs to qualitative data analysis.
Jul 15, 2026