Scaling DoRA: High-Rank Adaptation via Factored Norms and Fused Kernels
DoRA scaling improvements using factored norms and fused kernels to reduce memory overhead in weight-decomposed low-rank adaptation for LLMs.
DoRA scaling improvements using factored norms and fused kernels to reduce memory overhead in weight-decomposed low-rank adaptation for LLMs.
Off-topic: addresses passive torque control for robotic manipulators using viability theory for collision avoidance.
Introduces visual exclusivity attacks for multimodal models where harm emerges through visual content reasoning, exploited via agentic planning for red teaming.
Proposes fast-slow thinking reward models combining scalar and generative reward models for efficient RLHF alignment with improved accuracy over scalar-only approaches.
Presents AgenticGEO, a self-evolving agentic system for generative engine optimization that dynamically adapts content strategies to improve visibility in LLM-based search.
Proposes multi-agent debate with memory masking for LLM reasoning, where multiple agents debate solutions across rounds with selective memory management.
Introduces locally coherent parallel decoding for diffusion language models to capture token dependencies while achieving sub-linear generation latency.
Investigates predicting expected reward scores from reward models to route prompts to suitable LLMs before generation, enabling intelligent model selection.
Studies KV cache reuse strategies in chunk-level caching for retrieval-augmented generation, analyzing accuracy improvements when precomputing caches for retrieved text chunks.
Proposes latent lookahead training for transformers to enable multiple token exploration per step, addressing limitations of standard next-token prediction in autoregressive language models.
Compares latency and energy costs of edge vs cloud inference for AI tutoring using quantized Phi-3 models, analyzing learning-per-watt efficiency.
Convex relaxations for rank-constrained quadratic optimization without spectral structure requirements using lifted semidefinite programming.
Preordered Multi-Objective MDP for autonomous driving balancing safety, efficiency, and comfort via distributional reinforcement learning with safety constraints.
Driver risk fusion method for screening safety-critical scenarios in autonomous driving from naturalistic driving data without manual annotation.
Machine learning framework for severe weather prediction over 2-6 hour windows using convection-allowing model output post-processing.
Abjad-Kids dataset for Arabic speech recognition in primary education covering alphabets, numbers, and colors for low-resource language learning.
SciNav is an agent framework for autonomous science agents to perform scientific coding tasks with objective evaluation on executable benchmarks.
DESRO framework uses LLMs to infer intermediate scientific reasoning steps from experimental outcomes for molecule optimization without explicit step annotations.
Semisupervised geometric unmixing method using simplex-volume penalties and archetypal analysis for spectral data analysis.
Multi-agent reinforcement learning framework for coordinating UAV networks with joint communication, sensing, and energy constraints for waste detection.
Heterogeneous multi-agent reinforcement learning with learned inter-agent communication for autonomous cyber defense against network attacks.
Mix-and-Match Pruning globally-guided layer-wise sparsification framework for compressing DNNs with minimal accuracy loss on edge devices.
FastPFRec federated recommendation framework using GNNs with enhanced convergence speed and privacy-preserving secure aggregation.
Knowledge distillation approach mitigating catastrophic forgetting in incremental hyperspectral image classification without storing old samples.
VGS-Decoding training-free method to reduce hallucinations in medical vision-language models using visual grounding scores during inference.
Empirical study of how executable tool access impacts safety alignment in LLM agents. Shows tool affordance increases capability-safety misalignment.
Demonstrates mathematical isomorphism between ant colony decision-making and random forest ensemble learning under stochastic ensemble intelligence framework.
Mathematical study of forward and inverse problems for time-dependent probability measures in Bayes-Hilbert spaces.
Federated recommender system addressing subgraph structural imbalance in decentralized training on client-specific user-item graphs.
G2DR framework for genotype-first therapeutic target and drug discovery using genetic variants and transcriptomics data.
Theory for data-driven operator learning methods in smoothing and forecasting of dynamical systems and data assimilation.
Framework for representing and propagating measurement uncertainty beyond Gaussian assumptions in control and measurement systems.
Nexerra-R1 chemical language model for metal-organic framework design. Systematic discovery of materials with targeted properties via LLM.
Multi-agent optimization framework for resource allocation in heterogeneous LoRa IoT networks combining ground and underground sensors.
Introduces Meta-Persuasion algorithms applying meta-learning to repeated Bayesian persuasion games with theoretical guarantees.
CERN uses Hidden Markov Models to correct raw electrical signals from nanopore DNA sequencing for improved genome assembly.
Hawkeye system reproduces GPU matrix operations on CPU for verifiable ML without precision loss. Enables reproducible ML inference analysis.
Derives computable state-estimation error bounds for physics-informed neural network KKL observers in control systems.
Study evaluating detectability of LLM-assisted peer reviews. Tests five detectors on dataset simulating human-AI collaboration levels.
Feature attribution method for explaining machine learning decisions in ECG signal analysis with shift-invariant properties.
Goal-oriented learning of surrogate models for stochastic dynamical systems with error bounds on path-dependent observables.
Simulation-based inference framework enabling rapid neural network fitting across varying cognitive modeling assumptions and parameterizations.
Study finding language models report highest confidence when fabricating, with formal proof this is observational not capability limitation.
SC-Net operator learning framework for regularized inverse problems using spectral filtering with improved interpretability and generalization.
World model-based reinforcement learning approach for training Vision-Language-Action robotic models without costly real-world interaction.
Neural architecture incorporating five inductive biases for improved performance on tabular data compared to tree-based models.
Dense associative memory system for empirical measures using Sinkhorn divergence and spherical Hellinger Kantorovich gradient flows for pattern retrieval.
Framework for integrating structural and functional brain connectomes using hierarchical multiscale learning across nested modular organizations.
Asynchronous decomposition framework for high-dimensional online learning with dynamic regularization avoiding error bound divergence.
mmWave-Diffusion framework using conditional diffusion models for contactless respiration sensing with micromotion interference removal.