archive
Every paper Pith has read. Search by title, abstract, or pith.
7878 papers in cs.LG · page 1
-
The paper proposes RefDecoder
RefDecoder: Enhancing Visual Generation with Conditional Video Decoding
-
FutureSim shows top AI agents predict events at 25% accuracy
FutureSim: Replaying World Events to Evaluate Adaptive Agents
-
Tensor similarity algebraically checks when two networks compute the same function
When Are Two Networks the Same? Tensor Similarity for Mechanistic Interpretability
-
This paper introduces Shodh-MoE
Eradicating Negative Transfer in Multi-Physics Foundation Models via Sparse Mixture-of-Experts Routing
-
EHR tables sharpen timing in text-based clinical timelines
Text Knows What, Tables Know When: Clinical Timeline Reconstruction via Retrieval-Augmented Multimodal Alignment
-
Seamless blending cuts robot-hand takeover jitter by 99.8%
Hand-in-the-Loop: Improving Dexterous VLA via Seamless Interventional Correction
-
Memory model lets LLMs add knowledge without retraining
MeMo: Memory as a Model
-
RoSHAP gives stable rankings by summarizing SHAP distributions
RoSHAP: A Distributional Framework and Robust Metric for Stable Feature Attribution
-
Optimal logging policies minimize OPE error via reward-coverage balance
Logging Policy Design for Off-Policy Evaluation
-
Anomaly detection uncovers refinery LP errors and opportunities
From Data to Action: Accelerating Refinery Optimization with AI
-
AGOP from kernel regression recovers central subspace with fewer samples
Average Gradient Outer Product in kernel regression provably recovers the central subspace for multi-index models
-
SpeakerLLM turns speaker verification into natural-language reasoning
SpeakerLLM: A Speaker-Specialized Audio-LLM for Speaker Understanding and Verification Reasoning
-
Attention network cuts IRS pilots by 87 percent
Multi-Block Attention for Efficient Channel Estimation in IRS-Assisted mmWave MIMO
-
128 random demos suffice for strong RLVR results
Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance
-
DeepTokenEEG is a compact deep learning model that processes EEG brain signals using…
DeepTokenEEG Enhancing Mild Cognitive Impairment and Alzheimers Classification via Tokenized EEG Features
-
Neural emulators deliver real-time virtual circuits for tokamak control
Real-time virtual circuits for plasma shape control via neural network emulators
-
The paper introduces ICGPS, which uses meta-trained generative models for in-context…
In-Context Learning for Data-Driven Censored Inventory Control
-
Reinforcement learning estimates hidden states for multivariate HMM forecasts
DRL-STAF: A Deep Reinforcement Learning Framework for State-Aware Forecasting of Complex Multivariate Hidden Markov Processes
-
Learned potential reweights bridges to improve generative fidelity
Action-Inspired Generative Models
-
Neural solvers reach energy parity after 158000 deployments
An Amortized Efficiency Threshold for Comparing Neural and Heuristic Solvers in Combinatorial Optimization
-
Discriminant loss sharpens segmentation boundaries
Deep Image Segmentation via Discriminant Feature Learning
-
Min-Max-IRL reaches fast O(n^{-1}) rates without exploration
Fast Rates for Inverse Reinforcement Learning
-
Recursive models collapse internally before metrics detect it
Silent Collapse in Recursive Learning Systems
-
SAM worsens DRL backdoors while other fixes reduce them
Angel or Demon: Investigating the Plasticity Interventions' Impact on Backdoor Threats in Deep Reinforcement Learning
-
Neural corrections boost accuracy of implicit solvent models for proteins
All-atomistic Transferable Neural Potentials for Protein Solvation
-
Algorithm reduces Any-Order-PDIV runtime from millions of years to minutes
Woodelf++: A Fast and Unified Partial Dependence Plot Algorithm for Decision Tree Ensembles
-
Alignment lets robots predict tactile sensations from sight
Let Robots Feel Your Touch: Visuo-Tactile Cortical Alignment for Embodied Mirror Resonance
-
ML classifier beats rules at spotting BDD refactoring chances
Mining Subscenario Refactoring Opportunities in Behaviour-Driven Software Test Suites: ML Classifiers and LLM-Judge Baselines
-
Sequential feature recovery produces power-law scaling
Scaling Laws from Sequential Feature Recovery: A Solvable Hierarchical Model
-
Action tokens carry the training signal in agentic RL
Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy
-
Bandits recover multi-objective prompts more efficiently
Efficient Multi-objective Prompt Optimization via Pure-exploration Bandits
-
SeesawNet fuses normalized and raw sequences for better time series forecasts
SeesawNet: Towards Non-stationary Time Series Forecasting with Balanced Modeling of Common and Specific Dependencies
-
Single score reveals accuracy does not equal responsibility
Multi-Dimensional Model Integrity and Responsibility Assessment Index and Scoring Framework
-
Fine-tuning exposes physical direction in neural PDE weights
Discovering Physical Directions in Weight Space: Composing Neural PDE Experts
-
LLMs top out at 46 percent exact match on medication choices
RxEval: A Prescription-Level Benchmark for Evaluating LLM Medication Recommendation
-
Activation patching isolates how LLMs represent relative geography
Exploring Geographic Relative Space in Large Language Models through Activation Patching
-
LLM agents autonomously develop ML interatomic potentials
Lang2MLIP: End-to-End Language-to-Machine Learning Interatomic Potential Development with Autonomous Agentic Workflows
-
Product kernels recover saturation and multiple descent in high-dim KRR
Large Dimensional Kernel Ridge Regression: Extending to Product Kernels
-
Foldable layer norms convert exactly to faster RMSNorm
Enjoy Your Layer Normalization with the Computational Efficiency of RMSNorm
-
ArcGate activation adapts shape to raise remote sensing accuracy
ArcGate: Adaptive Arctangent Gated Activation
-
DRL agent cuts bike availability failures with one truck
Fully Dynamic Rebalancing in Dockless Bike-Sharing Systems via Deep Reinforcement Learning
-
Bi-level optimization automates data mixing in offline-to-online RL
ROAD: Adaptive Data Mixing for Offline-to-Online Reinforcement Learning via Bi-Level Optimization
-
A GNN-Transformer trained to imitate a lookahead heuristic selects reduced scenarios for…
Learning Scenario Reduction for Two-Stage Robust Optimization with Discrete Uncertainty
-
Schur projection keeps state-space neural nets stable
A Novel Schur-Decomposition-Based Weight Projection Method for Stable State-Space Neural-Network Architectures
-
Models evolve reusable skills library at test time
Test-Time Learning with an Evolving Library
-
Focused estimator improves PU learning on imbalanced data
Focused PU learning from imbalanced data
-
IIQ metric scores AI integration from 0 to 1000
Intelligence Impact Quotient (IIQ): A Framework for Measuring Organizational AI Impact
-
Guardrails adapt from sparse noisy failures via conservative induction
LiSA: Lifelong Safety Adaptation via Conservative Policy Induction
-
Orthogonal projection isolates hallucination signals in LLM answers
When Answers Stray from Questions: Hallucination Detection via Question-Answer Orthogonal Decomposition
-
Synthesized open-ended problems raise LLM coding scores by 8-12 points
FrontierSmith: Synthesizing Open-Ended Coding Problems at Scale