Tag: #research297 results
LITT introduces a Timing-Transformer architecture that aligns sequential events on a virtual relative timeline for event-timing-focused attention. It enables personalized clinical trajectory interpretations. Validated on EHR data from 3,276 breast cancer patients to predict cardiotoxicity onset.
LatentRxnFlow predicts reactions as continuous latent trajectories via Conditional Flow Matching from reactant-product pairs. Offers SOTA USPTO accuracy with trajectory diagnostics and uncertainty estimation. Enables error mitigation and reliable predictions.
AIvilization v0 deploys a resource-constrained artificial society with unified LLM agents. Features hierarchical planning, adaptive profiles, and human steering for long-horizon autonomy. Reproduces real market stylized facts like wealth stratification.
Language-Action Pre-training (LAP) represents robot actions in natural language for zero-shot transfer across embodiments without fine-tuning. LAP-3B, a 3B VLA, delivers over 50% success on novel robots and tasks. Enables efficient adaptation and unifies action prediction with VQA.
LakeMLB is a benchmark for machine learning in data lakes, focusing on multi-table union and join scenarios with real datasets from government, finance, and more. Supports pre-training, augmentation strategies. Evaluates tabular ML methods and releases datasets/code.
Online Causal Kalman Filtering models IS ratios as evolving latent states for stable RL in LLMs. Smooths noise while preserving token structure. Superior on math reasoning datasets.
KORAL integrates LLMs with Data and Literature Knowledge Graphs for SSD diagnostics from fragmented telemetry. Provides descriptive, predictive, prescriptive, what-if analysis with explainable insights. Outperforms expert methods on production traces.
Enhances LLM-guided evolution by evolving programs that propose, improve, and perturb solutions iteratively. Achieves new SOTA on hexagon packing and autocorrelation inequality benchmarks. Reduces LLM cognitive load via structured parameterization.
Hierarchical Zero-Order optimization decomposes network depth for efficient ZO in DNNs. Reduces query complexity from O(ML^2) to O(ML log L). Matches backpropagation accuracy on CIFAR-10 and ImageNet.
Presents experimental framework comparing human-led, AI-led, and hybrid vibe coding groups. Humans deliver superior iterative instructions, preventing AI-led performance collapse. Hybrids thrive with human direction and AI evaluation.