Research Feed Page 19

The paper introduces RAIL, a multi-agent framework that uses independent LLM experts to objectively evaluate the readiness level of AI projects and prevent maturity inflation.

SCULPT generates complex 3D objects by iteratively subtracting individual parts from a holistic 3D latent representation.

TabSOM introduces a mapping technique to transform tabular data into stable image representations that explicitly capture feature relationships to improve deep learning performance.

The paper demonstrates how to fine-tune open-weights models to shift from passive assistant behaviors to a proactive Socratic persona using targeted parameter-efficient techniques.

The authors derive a mathematical framework that models how neural network training converges using a small set of variables regardless of the model size.

The researchers developed a pedagogically constrained training environment to build models that learn only within a specific academic scope.

The RIFT framework enables robot policies to predict future states in a single pass, eliminating the need for iterative video generation during inference.

The paper demonstrates that reported language model awareness scores are heavily influenced by the specific framing of the prompt rather than intrinsic model properties.

The authors developed a three stage pipeline that integrates automated neural architecture search with hardware aware mapping to mitigate the performance penalties of INT4 quantization.

The paper demonstrates how to use bagging to create machine learning models that are resilient against adversarial examples while maintaining efficient training.

Researchers developed an AI agent named Faraday that automatically replicates ML and AI-for-science papers by inferring missing details and executing experiments within a secure containerized environment.

The paper introduces a doubly robust estimator using targeted regularization to correct sample selection bias when calculating conversion rates for clicked samples.

The paper introduces the Defensive Booster, an algorithm that combines two distinct performance goals for probabilistic forecasting into a single, efficient online system.

The paper introduces a collision avoidance system for robotic manipulators that adjusts velocity commands based on predicted future obstacle positions to improve safety in dynamic environments.

The researchers developed a task-agnostic method to quantify how individual training examples affect a language model's final parameters without needing to retrain the model.

The paper introduces UniTexture, a method that uses a single adversarial visual pattern to degrade the performance of multitask vision-language-action models across diverse robotics tasks.

The paper introduces a method called CROP to selectively focus model distillation on task-relevant information by measuring sensitivity to counterfactual prompts.

The paper introduces a protocol-level audit to determine if LLM benchmark scores represent genuine reasoning or are artifacts of confounding factors.

The researchers developed an automated pipeline that uses an LLM to generate graph-based representations of optimization problems, enabling efficient configuration selection for local search algorithms.

The paper provides a formal framework called Causal World Models that maps raw observations into structured latent variables to enable reliable causal decision making.