11d ago
CausalMix: Data Mixture as Causal Inference for Language Model Training
arXiv paper on CausalMix for optimizing data mixture weights in LLM training using causal inference without retraining.
11d ago
SynLaD: Latent Diffusion for Generating Synthesizable Molecules Conditioned on 3D Pharmacophore Profiles
arXiv paper on SynLaD latent diffusion framework for generating synthesizable drug molecules with pharmacophore constraints.
11d ago
Muon as a Residual Connection
arXiv paper proposing Muon optimizer as implicit residual connection during neural network training, explaining its effectiveness.
11d ago
ZO-Act: Efficient Zeroth-Order Fine-Tuning via One-Shot Activation-Informed Low-Rank Subspaces
arXiv paper presenting ZO-Act for efficient zeroth-order fine-tuning of large language models using activation-informed low-rank subspaces.
11d ago
GAIA: Geometry-Adaptive Operator Learning for Forward and Inverse Problems
arXiv paper on GAIA, a geometry-adaptive neural operator learning method for PDEs including boundary value and inverse problems.
11d ago
Sequentially-Controlled Interactive Multi-Particle Flow-Maps for Online Feedback-Driven Search
arXiv paper proposing IMPF for reward alignment in generative models with sequential feedback-driven exploration.
11d ago
A Lightweight Self-Supervised Learning Framework for Multivariate Time Series using Hierarchical-JEPA on ECG Data
ER-JEPA lightweight self-supervised learning framework using hierarchical joint-embedding predictive architecture for ECG signal analysis.
11d ago
Efficient Compression of Structured and Unstructured Volumes via Learned 3D Gaussian Representation
Learned 3D Gaussian representation for efficient compression of structured and unstructured volume data with reduced memory footprint.
11d ago
Decision-Aware Training for Sample-Based Generative Models
Decision-aware training framework for generative models in forecasting that optimizes for downstream decision maker costs instead of standard scoring rules.
11d ago
QuasiMoTTo: Quasi-Monte Carlo Test-Time Scaling
Quasi-Monte Carlo test-time scaling method for language models reducing redundancy in parallel sampling while improving inference efficiency.
11d ago
Right in the Right Way: LM Training with Verifiable Rewards and Human Demonstrations
RLVR framework combining verifiable rewards and human demonstrations for LM training, addressing diversity collapse from objective-only optimization.
11d ago
Neural Certificate Pricing for Combinatorial Optimization Problems
Neural Certificate Pricing applies unsupervised learning to combinatorial optimization by leveraging asymmetry between search and verification complexity.
11d ago
Quantum vs. Classical Machine Learning: A Unified Empirical Comparison
Empirical comparison of quantum machine learning models versus classical machine learning approaches across benchmarks.
11d ago
TiRex-2: Generalizing TiRex to Multivariate Data and Streaming
TiRex-2 extends univariate time series foundation model to multivariate forecasting using recurrent xLSTM with streaming capability.
11d ago
Language-Critique Imitation Learning from Suboptimal Demonstrations
Language-critique framework for imitation learning from suboptimal demonstrations using natural language feedback instead of scalar signals.
11d ago
Is One Layer Enough? Training A Single Transformer Layer Can Match Full-Parameter RL Training
Study showing single transformer layer training matches full-parameter RL fine-tuning for LLMs, revealing unequal layer-wise contribution during RL post-training.
11d ago
Why Advanced Encoders Lag on Sparse Retrieval? The Answer and an Approach to Bridging Vocabulary Gaps
Identifies vocabulary gap in modern encoders for sparse retrieval and proposes approach to bridge gap between dense and sparse retrieval.
11d ago
Urban Deceleration Behavior Modes Under Scene Context: An Early-Kinematic Classifier from Argoverse 2 Multi-Agent Trajectories
Kinematic classifier for urban vehicle deceleration behaviors using K-means clustering on Argoverse 2 trajectory data.
11d ago
Spatio-Temporal Gaussian Process for Building Terrain-Incorporating Wind Power Curves
Spatio-temporal Gaussian process model for wind turbine power curves incorporating terrain covariates.
11d ago
Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust
Framework using steering vectors and latent space analysis to control and calibrate language model behavior for trustworthy deployment.
11d ago
A Mechanism-Driven Theory of Phase Transitions in Active Learning
Theoretical characterization of active learning budget regimes as shifts in dominant generalization mechanisms.
11d ago
Steal the Patch Size: Adversarially Manipulate Vision-Language Models
Black-box attack recovering private vision-tokenizer configurations of vision-language models through side-channel analysis.
11d ago
HydraCollab: Adaptive Collaborative-Perception for Distributed Autonomous Systems
Multi-robot collaborative perception system balancing communication bandwidth and perception accuracy for autonomous systems.
11d ago
Homogenization of $\ell_2$-Adversarial Training in High-Dimensions: Exact Dynamics under Stochastic Gradient Descent
Theoretical framework analyzing adversarial training dynamics for single-index models on Gaussian mixtures using SGD in high dimensions.
11d ago
SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing
SLIM-RL proposes risk-budgeted random-masking reinforcement learning for diffusion LLMs, eliminating trajectory slicing overhead from prior TraceRL method.
11d ago
Sample Complexities of Estimating Gumbel--Max Watermark Proportions with and without Reduction to Pivotal Statistics
Analysis of sample complexity for estimating watermark proportions in documents under Gumbel-max LLM watermarking mechanism.
11d ago
Leveraging Multimodality for Real-Time Classification of Transients and Variables found by the Zwicky Transient Facility
Multimodal machine learning for real-time classification of transient astronomical objects from Zwicky Transient Facility survey.
11d ago
Computer vision-based neural networks for radioisotope identification in urban environments
Computer vision neural networks for radioisotope identification from gamma-ray spectrograms in urban environments.
11d ago
Self-Organized Learning in Oscillatory Neural Networks with Memristive Signed Couplings
Self-organized learning in oscillatory neural networks with memristive couplings for associative memory and optimization.
11d ago
Rosetta: Composable Native Multimodal Pretraining
Rosetta: Composable multimodal pretraining approach addressing gradient conflicts when integrating new modalities without catastrophic forgetting.
11d ago
DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning
DiscoLoop: Method for internalizing multi-hop reasoning in LLMs within single forward pass using discrete embeddings and continuous states.
11d ago
SoK: Attack and Defense Landscape of Mobile On-device AI Systems
Systematization of knowledge on attack and defense landscape for mobile on-device AI systems.
11d ago
The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models
Structured evaluation showing text-to-image diffusion safety alignment methods create illusion of high utility through coarse metrics.
11d ago
MindAU: EEG-Conditioned Facial Action Unit Editing via Dual-Stream Manifold Alignment
EEG-conditioned facial action unit editing using dual-stream manifold alignment.
11d ago
A Mechanistic View of Authority Hierarchy in LLM Sycophancy
Mechanistic investigation of authority bias in LLMs showing how models prioritize source credibility over factual consistency.
11d ago
Information-Regularized Attention for Visual-Centric Reasoning
Information-regularized attention mechanism to improve visual grounding and reduce hallucination in vision-language models.
11d ago
StochasT: Learning with Stochastic Turn Depth for Visual Instruction Tuning
StochasT: Visual instruction tuning method addressing visual attention decay in multi-turn vision-language model conversations.
11d ago
Predicting Lethal Outcome (Cause) And Understanding Key Biomarkers Linked With Acute Myocardial Infarction Using Deep Artificial Neural Network And Ensemble Of Machine Learning Methodologies
Deep neural networks and ensemble methods to predict mortality outcomes and identify biomarkers in acute myocardial infarction.
11d ago
How Environment and Urbanization Shape Bird Diversity in Sri Lanka
Analysis of bird diversity in Sri Lanka using spatial and environmental data.
11d ago
Optimal scaling of MCMC algorithms: exploiting the symmetry of the Metropolis-Hastings formula
Theoretical study of MCMC scaling properties using Metropolis-Hastings symmetry.
11d ago
Auditing Forgetting in Limited Memory Language Models
Causal auditing framework to detect whether deleted facts persist in limited memory language models through parametric memory or retrieval artifacts.
11d ago
Coachable agents for interactive gameplay
RL framework enabling interactive real-time control of agent behavior during gameplay through coachability mechanisms instead of learning single optimal policy.
11d ago
LUMA: Benchmarking Segmentation via a Lightweight Universal Mask Adapter
Lightweight backbone-agnostic segmentation benchmark adapter enabling fair comparison of transformer backbones independent of decoder and pretraining.
11d ago
Forensic-Oriented Intrusion Detection Using Synthetic Network Traffic Data and Explainable Artificial Intelligence
Intrusion detection framework using synthetic network traffic and explainable AI to produce court-defensible, traceable forensic outputs.
11d ago
Soft Mixture-of-Recursions: Going Deeper with Recursive Vision Transformers
Recursive Vision Transformer approach using soft mixture-of-recursions to build deeper models with better parameter efficiency and performance.
11d ago
Evaluating Pretrained Music Embeddings for Cross-Performance Jazz Standard Recognition
Evaluation of pretrained music embeddings for cross-performance jazz standard recognition, comparing CNNs against foundation model embeddings.
11d ago
Mirror-Fusion Attention for Reflection-Aware Self-Supervised Representation Learning
Self-supervised learning framework for Vision Transformers that preserves informative left-right correspondences in bilateral data like medical images.
11d ago
Shapley in Context: Explaining Financial Language with Domain Expertise
Method for explaining financial LLM decisions using Shapley values combined with domain expertise, addressing regulatory explainability requirements.
11d ago
MoVA: Learning Asymmetric Dual Projections for Modular Long Video-Text Alignment
Video-text alignment model using asymmetric dual projections to handle temporal misalignment and semantic heterogeneity in long videos.
11d ago
Graph-Native Reinforcement Learning Enables Traceable Scientific Hypothesis Generation through Conceptual Recombination
Graph-native RL system for materials discovery generating scientifically valid hypotheses through multi-step reasoning with traceable intermediate steps.