MoE-based LLM compression using SVD exploiting heterogeneous expert routing frequency and information density for practical deployment.
LLMs applied to designing participatory budgeting rules that optimize utility and fairness for democratic project funding allocation.
Uses learned LLM features as reward signals for reinforcement learning to reduce hallucinations and improve open-ended task performance.
Automated pipeline detects unverbalized biases hidden in LLM chain-of-thought reasoning without predefined categories.
Aletheia is a math research agent using LLMs for autonomous theorem discovery, proof generation, and literature navigation in mathematics research.
QTALE combines token-adaptive layer execution with quantization for efficient LLM deployment while handling robustness challenges.
Control Reinforcement Learning uses sparse autoencoders to identify and steer interpretable features at token-level for LLM steering and intervention.
LoopLMs improved via reinforcement learning that rewards intermediate reasoning steps rather than final outputs for better multi-step reasoning.
SnapMLA optimizes DeepSeek MLA decoding using FP8 quantization and hardware-aware pipelining for efficient long-context inference.
PRISM architecture balances transformer expressivity with linear model efficiency for sequence generation through parallel residual iterations.
Framework for compiling high-level neural network specifications into VNN-LIB format for formal verification of neural networks.
Technique to accelerate LLM inference by adding self-supervised early exit heads at intermediate transformer layers using confidence thresholds.
Framework using decision trees and optimization rules for interpretable surrogate models in optimization.
Benchmark evaluating safety of LLMs and vision language models on laboratory safety issues and procedural guidance tasks.
Dataset of 40,000 journalistic interviews to evaluate LLM grounding and strategic dialogue capabilities in informational interviews.
Model-free approach using reservoir computing for controlling dynamical systems to reach unseen target states.
Approximation algorithms for orthogonally constrained quadratic optimization using semidefinite programming techniques.
Framework training LLMs to generate policy explanations via reinforcement learning using generative normalizing flows for reward generation.
Research on robustness of retrieval-augmented generation systems, identifying and mitigating spurious features in grounding data that affect RAG performance.
EEG2GAIT: hierarchical graph convolutional network for decoding gait dynamics from EEG with multi-level spatial embeddings.
Theoretical study of learning in structured Stackelberg games with contextual information for leader-follower strategic interaction.
Framework for using LLMs to detect WebShell attacks with behavioral function-aware analysis, addressing data scarcity and concept drift.
GeneMamba: bidirectional Mamba architecture for single-cell RNA sequencing analysis with efficient context learning.
Domain randomization with model-based control for handling parametric uncertainty in nonlinear vehicle powertrain systems.
Theoretical framework for optimal probability density control on infinite-dimensional spaces with maximum principle and HJB equations.
BrainSymphony: parameter-efficient multimodal foundation model for brain dynamics combining fMRI and structural connectivity data.
NeuroDyGait: two-stage phase-aware framework for decoding lower-limb motion from EEG signals in brain-computer interfaces.
Topological data analysis framework for forecasting Overall Equipment Efficiency in Industry 4.0 manufacturing environments.
PhreshPhish: large-scale phishing website dataset and benchmark for evaluating ML-based phishing detection models.
Knowledge distillation method for training efficient facial animation models from speech without large diverse datasets.
Interpretability study examining how CNNs predict lexical stress in English words from speech data using neural network analysis.
MedQARo benchmark: 105,880 medical QA pairs in Romanian for evaluating LLM performance on medical question answering requiring reasoning.
SteerMoE framework for controlling Mixture-of-Experts LLMs by detecting and activating/deactivating behavior-associated expert networks.
Analysis of model collapse in generative models trained on their own synthetic outputs, deriving generalization error formulae for overparameterized linear regression.
Study of instrumental variable regression under differential privacy constraints, proposing noisy two-stage gradient descent algorithm.
Research on enhancing LLM reasoning by incorporating structured logical knowledge from training data through entailment relationships and logical complexity analysis.
Acoustic scene classification model using contrastive learning for edge devices that adapts to new acoustic categories without retraining.
Speech emotion recognition via generative data augmentation with mutual information regularization to improve consistency of synthetic emotional speech samples.
Economic theory research on tacit collusion between adaptive market agents in repeated games.
Research on artifact detection in generative image super-resolution weighted by human perceptual prominence.
LLM-powered framework (CELEC) for automated EHR data extraction and analytics via natural language SQL translation.
Research on symbolic regression using equality graphs to reduce search space for AI-driven scientific discovery.
Research on MapReduce LoRA and reward-aware training for multi-preference optimization in generative models.
Research analyzing stability properties of vector databases for retrieval-augmented generation and high-dimensional nearest-neighbor search.
Research on vision foundation models for medical image segmentation via uncertainty-informed collaborative learning.
Medical imaging research using generative modeling for pressure distribution synthesis.
Research on AI-driven super-resolution for encrypted point cloud video streaming in AR/VR.
Academic research on block-recurrent depth structure in Vision Transformers for mechanistic interpretation.
Three factor delay learning rules for spiking neural networks
Succeeding at Scale: Automated Dataset Construction and Query-Side Adaptation for Multi-Tenant Search