Every paper below was generated entirely by e5o — from a single topic prompt to a complete research manuscript with real experiments.
Investigates adaptive curriculum learning strategies on CIFAR-10/100 benchmarks. Proposes a difficulty-aware scheduling mechanism that dynamically adjusts training sample ordering to improve convergence speed and final accuracy.
Proposes a routing mechanism for Mixture-of-LoRA experts that considers prompt length characteristics. Fine-tunes Qwen-2.5-3B with QLoRA to demonstrate improved instruction-following across varying input lengths.
Extends graph attention networks with learnable edge feature transformations for molecular property prediction on the OGB-MolHIV benchmark, achieving competitive performance with existing specialized architectures.
Proposes entropy-guided intrinsic reward bonuses to improve exploration efficiency in sparse-reward MuJoCo locomotion environments. Demonstrates improved sample efficiency over baseline algorithms.
Systematically studies the effect of spectral normalization on mode collapse in conditional GANs trained on CIFAR-10, providing both visual and quantitative analysis (FID, IS) of generation diversity.
Explores lightweight test-time adaptation methods that update batch normalization statistics to handle distribution shift on CIFAR-10-C corruption benchmarks, demonstrating practical robustness improvements.
We're generating showcase papers across diverse ML subfields. Each paper will include a downloadable PDF, LaTeX source, experiment code, and quality assessment. Check back soon!