Machine Learning
Papers on learning algorithms, neural networks, deep learning, and optimization.
cs.LG · 1353 papersDense vs Sparse Pretraining at Tiny Scale: Active-Parameter vs Total-Parameter Matching
This paper compares dense and MoE transformers at tiny scale, finding MoE outperforms dense when matching active parameters but not total parameters.
High-Rate Quantized Matrix Multiplication II
This paper explores high-rate quantized matrix multiplication for LLMs, showing how waterfilling improves GPTQ and analyzing the near-optimal WaterSIC scheme.
VectorSmuggle: Steganographic Exfiltration in Embedding Stores and a Cryptographic Provenance Defense
VectorSmuggle reveals steganographic data exfiltration in RAG embedding stores and proposes VectorPin, a cryptographic defense for embedding integrity.
Min Generalized Sliced Gromov Wasserstein: A Scalable Path to Gromov Wasserstein
min-GSGW offers a scalable, rigid-motion invariant method for Gromov-Wasserstein by using generalized slicers and an amortized variant.
Robust and Explainable Bicuspid Aortic Valve Diagnosis Using Stacked Ensembles on Echocardiography
An explainable AI model accurately diagnoses bicuspid aortic valve (BAV) from tricuspid aortic valve (TAV) using routine echocardiography.
Children's English Reading Story Generation via Supervised Fine-Tuning of Compact LLMs with Controllable Difficulty and Safety
Fine-tuning compact 8B LLMs with expert curricula generates children's English stories with controllable difficulty and safety, outperforming larger models.
DisAgg: Distributed Aggregators for Efficient Secure Aggregation in Federated Learning
DisAgg uses distributed client aggregators to securely and efficiently aggregate updates in federated learning, achieving a 4.6x speedup over OPA.
Beyond Perplexity: A Geometric and Spectral Study of Low-Rank Pre-Training
Low-rank pre-training methods yield geometrically distinct solutions from full-rank models and each other, even with similar perplexity, requiring deeper evaluation metrics.
Multi-Objective and Mixed-Reward Reinforcement Learning via Reward-Decorrelated Policy Optimization
RDPO improves multi-objective and mixed-reward RL by decorrelating rewards and stabilizing advantage allocation for diverse reward types.
RealICU: Do LLM Agents Understand Long-Context ICU Data? A Benchmark Beyond Behavior Imitation
RealICU is a new benchmark for evaluating LLM agents on long-context ICU data, revealing recall-safety tradeoffs and anchoring biases in existing models.
Limits of Personalizing Differential Privacy Budgets
This paper reveals that personalized differential privacy budgets have significant limitations, showing a simple thresholding method is often superior.
Context-Aware Web Attack Detection in Open-Source SIEM Systems via MITRE ATT&CK-Enriched Behavioral Profiling
Smart-SIEM enhances open-source SIEMs with an AI module for context-aware web attack detection using behavioral profiling and MITRE ATT&CK.
Backdoor Channels Hidden in Latent Space: Cryptographic Undetectability in Modern Neural Networks
This paper shows how to create cryptographically undetectable backdoors in modern neural networks by exploiting latent space geometry, resisting current defenses.
LoREnc: Low-Rank Encryption for Securing Foundation Models and LoRA Adapters
LoREnc is a training-free framework that secures foundation models and LoRA adapters against IP leakage and model recovery attacks with minimal overhead.
Code-Centric Detection of Vulnerability-Fixing Commits: A Unified Benchmark and Empirical Study
This study finds code language models struggle to detect vulnerability-fixing commits without commit messages, lacking transferable security understanding from code changes alone.
Protocol-Driven Development: Governing Generated Software Through Invariants and Evidence
Protocol-Driven Development (PDD) governs generated software by using machine-enforceable protocols, invariants, and verifiable evidence chains.
A Resampling-Based Framework for Network Structure Learning in High-Dimensional Data
RSNet is an R package for robust, interpretable network inference in high-dimensional data, using resampling and graphlet analysis for structural insights.
scShapeBench: Discovering geometry from high dimensional scRNAseq data
scShapeBench introduces a benchmark and scReebTower, a new method for automated shape detection in high-dimensional scRNAseq data, outperforming baselines.
AlphaGRPO: Unlocking Self-Reflective Multimodal Generation in UMMs via Decompositional Verifiable Reward
AlphaGRPO enhances multimodal generation in UMMs using GRPO and a novel Decompositional Verifiable Reward for self-reflection and reasoning.
Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
Pion is a novel spectrum-preserving optimizer for LLMs that uses orthogonal transformations to maintain singular values throughout training.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.