archive
Every paper Pith has read. Search by title, abstract, or pith.
14903 papers in cs.LG · page 5
-
Fibonacci ring aggregation outperforms FedAvg in federated learning
FIRMA: FIbonacci Ring Model Aggregation for Privacy-preserving Federated Learning
-
Sparse autoencoder links reasoning steps to image masks
SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation
-
Timed precursor lifts secretary success above 50 percent
The Secretary Problem with a Stochastic Precursor
-
Causal model matches age changes in spine DXA images
From Baseline to Follow-Up: Counterfactual Spine DXA Image Synthesis in UK Biobank Using a Causal Hierarchical Variational Autoencoder
-
SGD variance grows unbounded along flat directions
Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics
-
Moral knowledge retrieval beats extra context for political value detection
More Context, Larger Models, or Moral Knowledge? A Systematic Study of Schwartz Value Detection in Political Texts
-
Moral knowledge beats extra context and model scaling for value detection
More Context, Larger Models, or Moral Knowledge? A Systematic Study of Schwartz Value Detection in Political Texts
-
CAME-Grad optimizer lifts radiology reports by 2 percent
The Double Dilemma in Multi-Task Radiology Report Generation: A Gradient Dynamics Analysis and Solution
-
CAME-Grad fixes gradient double dilemma in report generation
The Double Dilemma in Multi-Task Radiology Report Generation: A Gradient Dynamics Analysis and Solution
-
Frozen LLM corrections improve predictions within but not across protocols
From Residuals to Reasons: LLM-Guided Mechanism Inference from Tabular Data
-
WPO converges linearly to optimum under entropy regularization
A note on convergence of Wasserstein policy optimization
-
Hybrid detector catches unseen network attacks above 98% F1
UNAD+: An Explainable Hybrid Framework for Unknown Network Attack Detection
-
Dual rewards stabilize unsupervised LLM reasoning
Two is better than one: A Collapse-free Multi-Reward RLIF Training Framework
-
Shared program evolution then adaptation beats single-task search
Evolutionary Multi-Task Optimization for LLM-Guided Program Discovery
-
Healthcare LLM benchmarks fail because of hidden user assumptions
Healthcare LLM Benchmarks Are Only as Good as Their Explicit Assumptions
-
Data characteristics drive ML performance in PICU stewardship
Benchmarking Machine Learning Architectures for Antimicrobial Stewardship in Pediatric ICUs
-
Agentic-VLA speeds VLA convergence 2.4x with adaptive rewards
Agentic-VLA: Efficient Online Adaptation for Vision-Language-Action Models
-
AI Framework Secures Cardless Banking Against Fraud
Innovations in Cardless Artificial Intelligence Banking: A Comprehensive Framework for Cyber Secure and Fraud Mitigation using Machine Learning Algorithms
-
Residual stress learning narrows real-to-sim gap in dynamics
MoSA: Motion-constrained Stress Adaptation for Mitigating Real-to-Sim Gap in Continuum Dynamics via Learning Residual Anisotropy
-
Single network generalizes robot control to new factor mixes
Factored Diffusion Policies:Compositionally Generalized Robot Control with a Single Score Network
-
Ensembles add little uncertainty value for graph neural networks
Do Deep Ensembles Actually Capture Uncertainty in Graph Neural Networks?
-
Noise prediction loss matches score matching up to constant
A Tutorial on Diffusion Theory: From Differential Equations to Diffusion Models
-
3D reconstruction turns floorplan localization into alignment task
SceneAligner: 3D-Grounded Floorplan Localization in the Wild
-
Graph of atomic ops boosts LLM agent accuracy and cuts memory 4x
GraphFlow: A Graph-Based Workflow Management for Efficient LLM-Agent Serving
-
Multiple metrics required to judge synthetic data for tool-calling agents
SynAE: A Framework for Measuring the Quality of Synthetic Data for Tool-Calling Agent Evaluations
-
Tighter regret bounds let BO stop with optimality guarantees
Regret-Based $(\epsilon,\delta)$-optimal Stopping Criteria for Bayesian Optimization
-
Neural flows approximate any operator on function spaces
Neural Flow Operators can Approximate any Operator: Abstract Frameworks and Universal Approximations
-
Wavelet-guided neural terrain models reach 66 dB PSNR
ImplicitTerrainV2: Wavelet-Guided Spatially Adaptive Neural Terrain Representation
-
Martingale kernel tests replace permutations with normal quantiles
A Martingale Kernel Independence Test
-
Filtered sampling lets diverse models train together in GRPO
F-TIS: Harnessing Diverse Models in Collaborative GRPO
-
Linear maps predict object embeddings from subject embeddings
Relational Linear Properties in Language Models: An Empirical Investigation
-
RICA defines local disentanglement with a Hessian-Ricci tensor
Disentanglement Beyond Generative Models with Riemannian ICA
5 Piths -
Multicollinearity inflates AI explanation variance in cybersecurity
Stabilising Explainability Fragility in Cybersecurity AI: The Impact and Mitigation of Multicollinearity in Public Benchmark Datasets
-
Joint token diffusion policy scales language humanoid control
SCRIPT: Scalable Diffusion Policy with Multi-stage Training for Language-driven Physics-based Humanoid Control
-
New EEG dataset benchmarks meditation state and technique classification
L-FAME: Longitudinal Focused Attention Meditation EEG Dataset and Benchmark
-
TabPFN lags behind GLM and XGBoost in insurance pricing tests
Is TabPFN the Silver Bullet for Insurance Pricing?
-
Value functions create straight paths for generative transport
Generative Modeling by Value-Driven Transport
-
Benign references anchor clustering to filter variable poisoning
EnCAgg: Enhanced Clustering Aggregation for Robust Federated Learning against Dynamic Model Poisoning
-
Workflows baked into small model weights cut agent costs 100x
Compiling Agentic Workflows into LLM Weights: Near-Frontier Quality at Two Orders of Magnitude Less Cost
-
Compiler turns programs into exact neural modules
The Neural Compiler: Program-to-Network Translation for Hybrid Scientific Machine Learning
-
Flows detect OOD via atypical latent noise
The Signal in the Noise: OOD Detection Through Goodness-of-Fit Testing in Factorised Latent Spaces
-
Multimodal policies fail differently depending on latent or generative setup
Understanding Multimodal Failure in Action-Chunking Behavioral Cloning
-
Transformer represents arithmetic intermediates without causal use
Represented Is Not Computed: A Causal Test of Candidate Algorithmic Intermediates in a Transformer
-
Stronger backdoor triggers can raise clean accuracy in high dimensions
When Stronger Triggers Backfire: A High-Dimensional Theory of Backdoor Attacks
5 Piths -
Random node sampling matches full GNN training on most datasets
Implicit Regularization of Mini-Batch Training in Graph Neural Networks
-
Blockwise resolvent attention runs entity tracking in O(n to 4/3 d) time
Structured-Sparse Attention for Entity Tracking with Subquadratic Sequence Complexity
-
WTA bottleneck forces symbolic feature encodings
Winner-Take-All bottlenecks enforce disentangled symbolic representations in multi-task learning
-
Graph tokenization fixes transformer depth for structure recovery
Lost in Tokenization: Fundamental Trade-offs in Graph Tokenization for Transformers
5 Piths -
Point estimators narrow spectra in multimodal inverse problems
Pointwise Metrics Mislead: An Evaluation Protocol for Multimodal Inverse Problems
-
Spectral alignment boosts cross-subject F1 in biomedical signals
BioFormer: Rethinking Cross-Subject Generalization via Spectral Structural Alignment in Biomedical Time-Series