Neural ensemble Kalman filter: Data assimilation for compressible flows with shocks
Neural ensemble Kalman filter for data assimilation in compressible flows with shocks, addressing spurious oscillations from bimodal distributions.
Neural ensemble Kalman filter for data assimilation in compressible flows with shocks, addressing spurious oscillations from bimodal distributions.
Evaluates interpretability of ML models on polynomial root classification benchmark, testing decision trees, regression, and neural networks for structure recovery.
Introduces ANTShapes, a simulator for neuromorphic vision datasets to address scarcity of Dynamic Vision Sensor data for anomaly detection research.
Improves Laplace mechanism for differentially private SGD in high-dimensional models using majorization theory, applicable to LLM fine-tuning.
Framework using auxiliary-variable-guided generative models to disentangle physical factors in dark matter halo representations.
Topological method for reachability analysis in robotics using Morse graphs to estimate regions of attraction in learned latent spaces.
Few-shot continual learning approach for 3D brain MRI using frozen foundation models with task-specific LoRA modules for tumor segmentation and brain age estimation.
Theoretical study of partition function estimation under bounded f-divergence with information-theoretic analysis.
Applies reinforcement learning to automated PCB component placement, addressing challenges like component size variation and wirelength constraints.
Completeness and bounding results for causal identification using counterfactual data from Layer 3 of Pearl's Causal Hierarchy.
Probabilistic framework for symbolic regression using variational inference over soft symbolic trees for scientific discovery with uncertainty quantification.
Neural radiance field method incorporating uncertainty quantification capturing both aleatoric and epistemic uncertainty for 3D scene reconstruction.
Hybrid reinforcement learning approach (RL-CMSA) for solving min-max multiple traveling salesman problems with iterative construction and adaptation.
Efficient image captioning via hyperdimensional cross-modal alignment of frozen language and vision models without multimodal fine-tuning.
Causal discovery method distinguishing whether variables influence the mean versus variance of other variables in heteroscedastic data.
Retrieval system learning node-specific Riemannian metrics for geometry-aware semantic search on citation graphs.
Pool-based active learning approach for exoplanet habitability classification under extreme class imbalance.
General Bayes framework for policy learning where decision rules are the target rather than outcome prediction.
Platform for standardized access to remote sensing foundation model embeddings across heterogeneous model formats and interfaces.
Neural operator method using boundary integral formulation for efficient mesh deformation in computational geometry.
Dynamic benchmarking protocol where AI agents autonomously generate, validate, and solve problems to evaluate LLM reasoning capabilities beyond static datasets.
Non-parametric crime prediction model using spatial hotspot mapping for law enforcement resource allocation.
Framework for causal discovery in longitudinal systems accounting for institutional workflow constraints and partial ordering of operational data.
Open-source interpretability tool for analyzing gated activation functions (SwiGLU) in transformer neurons across recent language models.
Addresses catastrophic forgetting in continual fine-tuning of LLMs for vulnerability detection in source code, using selective replay with LoRA on temporal distribution shifts.
MI²DAS: multi-layer intrusion detection framework for IIoT with incremental learning to detect novel attacks in dynamic environments.
Distributed optimization algorithm combining semismooth Newton method with augmented Lagrangian for network-based collaborative learning.
Vehicle-to-Everything (V2X) communication approach for cloud-assisted perception in autonomous driving under latency constraints.
Benchmark study evaluating cross-domain transferability of flow-based feature sets across IoT and IIoT datasets for intrusion detection.
RF-Agent: automated reward function design for reinforcement learning using LLM-based tree search to optimize low-level control tasks.
BUSD-Agent: cascaded multi-agent framework for breast ultrasound screening and diagnosis reducing biopsy referrals through selective decision-making.
SegMate: lightweight 2.5D architecture for medical image segmentation achieving state-of-the-art accuracy with reduced computational requirements.
Autonomous robotic assembly framework using reinforcement learning for constructing stable structures without predefined plans.
Benchmarking 10 BERT variants for Nepali sentence-level topic classification, evaluating multilingual and Indic-specific models.
Inference-time optimization for protein ensemble generation using experiment-guided diffusion with thermodynamic constraints.
Jailbreak Foundry: multi-agent system translating jailbreak papers into executable modules for unified benchmarking and reproducible LLM robustness evaluation.
Data-driven optimization pipeline for GPU efficiency in distributed LLM adapter serving, maximizing throughput with concurrent adapter hosting.
Two-stage unsupervised pipeline for IoT device traffic profiling with incremental model adaptation using density-based clustering.
Analysis of monoculture in LLMs showing agreement metrics depend on subjective baseline assumptions for independence.
DiffusionHarmonizer: bridges neural reconstruction and photorealistic simulation for autonomous robots using online diffusion enhancement.
Artificial Agency Program: research agenda for building resource-bounded AI agents driven by curiosity-as-learning-progress and human-tool integration.
RAViT: multi-branch vision transformer framework reducing computational cost for image classification through resolution-adaptive processing.
Automated auditing framework for detecting systematic failures in medical image classifiers using multimodal features and slice discovery.
Proposes uncertainty quantification method for multimodal LLMs using incoherence-adjusted semantic volume for reliable deployment.
SenCache: training-free acceleration for diffusion model inference via sensitivity-aware caching of model outputs across timesteps.
MuViT: multi-resolution vision transformers for analyzing gigapixel microscopy images across multiple spatial scales.
BLISSNet: deep operator learning model for fast, accurate fluid flow reconstruction from sparse sensor measurements.
Presents learning-augmented algorithms for approximate minimum spanning tree via metric forest completion with improved approximation bounds.
FaultXformer: transformer encoder for fault classification and location in electrical distribution systems using PMU data.
Formalizes desiderata showing compositional generalization requires linear, orthogonal representations in vision embedding models.