Physics-informed PointNets and geometry-aware neural operators for modeling flows across porous structures with coupled physics and diverse geometries.
Sanity checks validating whether sparse autoencoders recover meaningful features beyond random baselines for neural network interpretability.
ROAST uses on-distribution rollouts for parameter-efficient LLM activation steering at inference time, replacing off-distribution supervision with continuous soft scaling.
Plug-and-play regularization losses for Mixture-of-Experts models promoting expert specialization across intra- and cross-layers without structural modifications.
Comprehensive analysis of malicious prompt classifier robustness under distribution shift with 18 datasets spanning jailbreaks and prompt injections for LLM agents.
Pivot-driven resampling technique for deep dense exploration in LLM RL, discovering high-quality trajectories within limited sampling budget from language space.
TS-Haystack benchmark evaluates time series language models on long-context retrieval with millions of datapoints, requiring precise temporal localization.
Characterizes optimal batch size scheduling for large-scale deep learning under fixed data budget using functional scaling law framework.
MAGE optimizes KV cache memory access in block diffusion LLMs for long-context settings using dynamic sparse attention adapted to block diffusion uniqueness.
RMB-CLE framework for multi-task learning integrating error-based task clustering with local ensembling to mitigate negative transfer from unrelated tasks.
Analysis identifying five recurring biases in financial LLM applications: look-ahead, survivorship, narrative, objective, and cost bias that invalidate deployment claims.
MAD framework treats tabular anomaly detection as multi-agent debate, leveraging disagreement from heterogeneous model families under distribution shift and rare-anomaly regimes.
Transfer learning approach using LSTM for cross-household hot water demand forecasting to optimize heat pump operation and reduce energy waste.
Radial-VCReg augments VCReg with radial Gaussianization loss for improved self-supervised representation learning by aligning feature norms with Chi distribution.
Framework leveraging transformer-based language models for causal inference from unstructured text, comparing estimates against structured data baselines.
Testing methodology for AI/ML and quantum systems addressing high-dimensional inputs, probabilistic outputs, and evaluation of trustworthiness, fairness, and robustness.
Framework for adaptive multi-turn LLM interactions to efficiently elicit group-level information from surveys, optimizing respondent selection and questioning strategy.
KernelBlaster uses agentic workflows with in-context RL to optimize CUDA code across GPU architectures, aggregating knowledge from prior optimizations without expensive finetuning.
MLAT framework exposes pre-trained ML models as callable tools within LLM agent workflows, enabling agents to invoke quantitative predictions and reason about outputs contextually.
Federated learning approach (DeepFusion) for training MoE-based LLMs using knowledge distillation from heterogeneous edge devices, enabling privacy-preserving distributed training.
Applies conformal Signal Temporal Logic (STL) specifications to enhance safety and robustness of RL control in aerospace (F-16 simulation), encoding control objectives formally.
Adaptive efficient rollout optimization (Train Less, Learn More) for Group Relative Policy Optimization in LLM post-training, reducing redundant rollouts when group outcomes are identical.
Zero-shot instruction following in multi-task RL using linear temporal logic (LTL) representations to specify temporally extended tasks for generalist agent policies.
Multi-class online fuzzy classifier for dynamic environments with human-defined antecedent fuzzy sets and learned consequent values in streaming data settings.
Information-theoretic framework explaining data augmentation's role in generalization and invariance learning, providing theoretical justification for augmentation effectiveness.
Framework for evaluating robustness of ML interpretability methods (LIME, SHAP) in hydrocarbon prospect risking using geophysical tabular data classification.
Addresses activation outliers in transformer quantization through spectral decay technique (S2D), establishing correlation between pre-training scale and outlier severity with theoretical analysis.
Framework analyzing reasoning modalities (code, natural language, hybrid) in LLMs under token constraints, evaluating performance tradeoffs for reasoning-specialized models.
Novel attention mechanism (SSA) replacing dot-product self-attention with Kuramoto model solution, reducing quadratic complexity and grounding in biological neural computation.
Training-free activation sparsity method (WiSparse) for efficient LLM inference considering weight-aware interactions and inter-block sensitivity, reducing computation and memory access.
Multi-agent collaboration framework for discovering latent causal variables, overcoming limitations of traditional causal discovery algorithms that assume no latent confounders.
Identifies 'silent inconsistency' in data-parallel fine-tuning of LLMs where worker-level optimization dynamics misalign despite synchronized parameters, impacting training quality.
Reinforcement learning approach (LACONIC) for controlling LLM response length during training without fixed heuristic reward shaping, addressing inference latency and computational overhead.
Multi-armed bandit algorithm (SOAR) for heterogeneous noise sources that adaptively selects data sources to minimize regret, applicable to federated or multi-source learning scenarios.
Novel parameter-efficient fine-tuning method for LLMs using mixture of experts in alternative geometric spaces (hyperbolic, spherical) to capture complex language data structures.
Theoretical analysis using numerical methods to explain why GLU variants scale better than MLPs in frontier LLMs, grounding empirical architectural choices in function approximation theory.
Open-source framework for multi-task learning to rank using transformers and self-attention for multiple relevance criteria.
Constrained MAB algorithm for non-stationary environments with unknown constraints under adversarial and stochastic settings.
Economic framework for auditing machine unlearning compliance with regulatory data deletion requirements.
Framework for multi-agent RL with dynamic agent creation and reproduction, extending MARL beyond fixed agent counts.
Open dataset and benchmarks for ML-based drift chamber track reconstruction with GNNs.
Semi-structured sparsity method (N:M) for training deep RL agents from scratch with hardware acceleration.
Study of algorithmic replicability in constrained multi-armed bandit problems for reproducible ML experiments.
Continuous-time reinforcement learning method using Hamiltonian flow for event-driven control problems.
Graph learning benchmark dataset for evaluating methods on opioid crisis prediction and intervention.
Information bottleneck regularizer for concept bottleneck models to improve interpretability while maintaining accuracy.
Lightweight adapter aligns compressed DL model embeddings with original models to improve performance in resource-constrained deployment.
Optimization method for orthogonal matrix constraints in machine learning, improving upon Landing algorithm for scalability.
Analyzes and corrects diversity bias in deep generative models by comparing sample diversity to underlying data distribution.
SynthSAEBench toolkit for large-scale synthetic benchmarking of sparse autoencoders with realistic feature characteristics.