Is RL Dead in LLM Post-Training? MIT’s RandOpt Challenges Traditional Methods

The MIT‑CSAIL paper introduces RandOpt, a single‑step, gradient‑free, fully parallel post‑training algorithm that adds Gaussian noise to pretrained LLM weights and ensembles the results, achieving or surpassing PPO/GRPO performance by exploiting dense "neural thickets" that emerge as model scale grows.

LLMRandOptReinforcement Learning

0 likes · 12 min read

Is RL Dead in LLM Post-Training? MIT’s RandOpt Challenges Traditional Methods

Meituan Technology Team

Jun 9, 2022 · Artificial Intelligence

FSL++: A Few-Shot Learning Model for Chinese Language Understanding that Tops the FewCLUE Benchmark

FSL++—a RoBERTa‑large‑based few‑shot model enhanced with domain‑adaptive pre‑training, prompt learning, diverse embedding‑level augmentations, and ensemble self‑training—topped the Chinese FewCLUE benchmark, beating human accuracy on news and scientific classification tasks and delivering measurable gains across multiple Meituan product scenarios.

Chinese language understandingFew‑Shot LearningNLP

0 likes · 23 min read

FSL++: A Few-Shot Learning Model for Chinese Language Understanding that Tops the FewCLUE Benchmark

360 Tech Engineering

May 21, 2019 · Artificial Intelligence

Understanding Residual Networks: Ideas, Mechanisms, Variants, and Insights

This article reviews the concept of residual networks, explains their working principle and data‑flow interpretation, discusses why they improve deep models, analyzes path‑length effects on gradients, and surveys various residual block designs and practical takeaways.

Model architectureNeural NetworksResNet

0 likes · 9 min read

Understanding Residual Networks: Ideas, Mechanisms, Variants, and Insights

Xianyu Technology

May 10, 2018 · Artificial Intelligence

Mercari Price Prediction Using TFIDF, GRU, and Ensemble Models

By converting Mercari’s product titles, descriptions, and categorical data into TF‑IDF vectors and embeddings, training MLP and GRU networks, and ensembling them with weighted averaging, the authors achieve a 0.3873 RMSLE—matching the competition’s top score—and demonstrate the power of text‑only price prediction for C2C marketplaces.

GRUTFIDFensemble

0 likes · 8 min read

Mercari Price Prediction Using TFIDF, GRU, and Ensemble Models

21CTO

Sep 20, 2017 · Big Data

Winning O2O Coupon Redemption with XGBoost, GBDT, and Feature Engineering

This article details a data-driven solution for the 2016 O2O coupon redemption competition, describing dataset partitioning, extensive feature engineering across user, merchant, and coupon dimensions, handling leakage, and model fusion using XGBoost, GBDT, and RandomForest, achieving top AUC scores through weighted ensemble.

GBDTXGBoostcoupon redemption

0 likes · 12 min read

Winning O2O Coupon Redemption with XGBoost, GBDT, and Feature Engineering