All Articles
17524 articles total
Latent graph encoding of multimodal neuroimaging features with generative AI architectures
Development of gMMVAE, a multimodal graph variational autoencoder for encoding neuroimaging features from MRI data.
Ad Headline Generation using Self-Critical Masked Language Model
Proposes a programmatic solution for generating e-commerce ad headlines using Reinforcement Learning Policy gradient methods on Transformer-based Masked Language Models.
Gradient-Based Speech-to-Text Alignment for Any ASR Model: From CTC to Speech LLMs
Introduces a training-free, gradient-based method for speech-to-text alignment that works across diverse ASR models, including speech LLMs.
A Gold-Standard Study of What Makes a Lightweight Game-Playing Agent Strong
Presents a lightweight, game-agnostic recipe for training competitive RL agents for imperfect-information card games, released as a reusable package.
GemNav: Discrete-Token Visual Robot Navigation using a Multimodal Large Language Model
Introduces GemNav, a data-efficient robot navigation policy that adapts a frozen MLLM using LoRA on the language tower without auxiliary visual encoders.
ReMoDEx: A Local-to-Global Relevance-Based Model Decision Explainability Framework for large-Scale Image Datasets
Proposes ReMoDEx, a scalable framework for assessing model decision behavior in image classification to detect shortcut learning via relevance maps.
Computing with Stochastic Oracles in AI-Augmented Computation
Studies the Stochastic-Oracle Turing Machine (SOTM) framework to model AI-augmented computation and analyze the impact of response reuse on performance.
LoCA: Spatially-Aware Low-Rank Convolutional Adaptation of Vision Foundation Models
Introduces LoCA, a parameter-efficient fine-tuning framework that decouples channel and spatial adaptation for Vision Foundation Models.
MADB: A Large-Scale Music Aesthetics Dataset with Professional and Multi-Dimensional Annotations
Introduces MADB, a large-scale dataset for music aesthetics assessment with professional multi-dimensional annotations.
Imputation Meets Clustering: Exploiting Latent Subgroup Structure for Missing Data Recovery
Proposes CAGI, a framework that co-optimizes clustering and imputation to recover missing data by exploiting latent subgroup structures.
Comprehensive Evaluation of Large Language Model Responses: A Multi-Factor Scoring System
Presents a multi-factor scoring system for the comprehensive evaluation of LLM responses, including accuracy, conciseness, and factual consistency.
Postgres rewritten in Rust, now passing 100% of the Postgres regression tests
A project has rewritten the PostgreSQL database engine in Rust, successfully passing 100% of the official regression tests.
How Donkey Kong Toppled Atari
An exploration of the historical competition between Donkey Kong and Atari during the early era of gaming.
I Built the Only 2026 WWII Jeep
A detailed account of building a replica of a rare 1942 WWII Jeep.
SmartHomeSecure: Automated Detection and Repair of Smart Home Configuration Errors Using Large Language Models
SmartHomeSecure uses program analysis and constrained LLM generation to automatically detect and repair configuration errors in Home Assistant YAML files.
AirPASS: Over-the-Air Federated Learning via Pinching Antenna Systems
AirPASS introduces an alternating optimization framework for over-the-air federated learning using pinching antenna systems to improve wireless device selection and beamforming.
From Agentic to Autogenic Network Management for AI-Native 6G and Beyond: A Standards Perspective
A proposal for 'Autogenic' network management in 6G, enabling systems to self-program and evolve their own automation software using AI agents.
Enhancing deep learning models for time series classification via knowledge distillation
Research on applying knowledge distillation to make deep learning models for time series classification more efficient for resource-limited environments.
What Predicts Correctness in Text-to-SQL? A Selective-Prediction Study
A study on predicting the correctness of Text-to-SQL queries, finding that LLM-based verification and ensembles outperform self-consistency and log-probability signals.
A Multi-Analyst LLM Pipeline for Auditable Rule Discovery Across 68 Public Physiological Corpora
An LLM-driven pipeline for extracting and auditing rule shapes from 68 public physiological corpora to guide the design of contactless monitoring hardware.