Research Feed

DiffusionGemma improves language model speed by generating large blocks of text simultaneously instead of writing one word at a time.

The researchers developed a new system called AquaJEPA that allows underwater robots to accurately predict their future movement and surroundings even when camera or sonar sensors provide incomplete information.

The researchers developed a new scoring method called the Explainability Performance Coefficient to objectively measure how accurately machine learning models explain their decision making processes.

TFGformer is a new forecasting model that combines graph-based patterns with external information to predict future trends more accurately.

The researchers developed a new system that uses analytical tools and multiple evaluation goals to help language models discover more accurate scientific formulas.

The paper introduces ModelEquivBench, a new evaluation system that provides independent, verifiable proof of whether optimization models generated by large language models are actually correct.

The researchers developed a new system that reduces the massive computational effort required to find effective quantum computer circuit designs by predicting results instead of testing every single option.

Researchers studied how reducing the precision of translation model data affects speed and accuracy when processing long documents.

The researchers developed a method that allows autonomous cars to adjust their driving behavior based on a continuous scale of aggressiveness to better match user preferences.

The researchers created a system called AgenticRepair that gathers diverse types of background information about code to help AI agents fix security flaws more reliably.

The researchers developed a new classification system that identifies exactly which part of an artificial intelligence agent caused a mistake, making it easier to fix errors.

The researchers developed a new system called MolGVR that improves how artificial intelligence converts written chemical descriptions into accurate molecular structures by using a three part verification and correction process.

The paper investigates if training language models on data that shows cause and effect improves their ability to reason correctly in complex, misleading scenarios.

The paper demonstrates that neural networks reaching high-entropy equilibrium states are more resistant to forgetting previously learned information when forced to memorize new data.

The researchers created a new benchmarking tool called DungeonBench to measure how well artificial intelligence models navigate the complex combat rules and resource management required in Dungeons and Dragons.

This research introduces a method for robots to generate movement plans by starting from their recent history rather than random noise, making the process faster and more efficient.

The paper investigates why AI agents struggle to act on user information they have already remembered by measuring the gap between recall and behavioral application.

The paper introduces DivAlign, a system designed to prevent AI tools from suggesting the same repetitive research directions to different scientists.

TerraNova is a foundation model that integrates continuous environmental data with discrete human administrative records to better understand their interaction.

The paper introduces a new method to reduce energy consumption in edge devices by balancing how neural networks are split and compressed during collaborative learning.