Seoul National University & Adobe Present: Program Synthesis via Test-Time Transduction A formulation that explicitly leverages test inputs during synthesis, unlike prior approaches that only generalize from training examples. Key idea: treat synthesis as active learning over a
MACHINE LEARNING
-
CAKE: Adaptive Kernel Design for Bayesian Optimization
By
–
Adaptive Kernel Design for Bayesian Optimization Is a Piece of CAKE with LLMs The Chinese University of Hong Kong, Shenzhen, and others
Paper: https://
alphaxiv.org/abs/2509.17998
v2
…
Code: https://
github.com/richardcsuwand
i/cake
… -
CAKE: Context-Aware Kernel Evolution for Bayesian Optimization
By
–
Is your Bayesian Optimization slow to converge? You might be using the wrong recipe! This NeurIPS 2025 paper introduces CAKE , Context-Aware Kernel Evolution, a freshly baked method that uses Large Language Models (LLMs) as genetic operators to adaptively evolve the perfect
-
RecoWorld and Meta Modern Recommendation System for Agentic AI
By
–
#PapersAccepted by Jiqizhixin
Our report: https://
mp.weixin.qq.com/s/J2qoNk3UCRlT
1wHLsvOHTA
… RecoWorld: Building Simulated Environments for Agentic Recommender Systems Meta Modern Recommendation System (MRS)
Paper: https://
arxiv.org/abs/2509.10397 -
New Book Builds on LLM from Scratch with Reasoning Techniques
By
–
Yes! It basically picks up where the LLM from Scratch left of. I.e., we start with a pre-training LLM and then implement and add reasoning techniques.
(However, it can also be read as a standalone book) -
Learning Bits Through RL and SFT: Research Insights
By
–
best paper or blog i've read in a while, highly recommend! John is brilliant and his research sets an example for the rest of us. recently i too have been thinking deeply about how many bits might be learned via one step of RL or SFT.. if you're thinking about this too, lmk!
-
PromptCoT 2.0: Scaling Prompt Synthesis for LLM Reasoning
By
–
PromptCoT 2.0 Scaling Prompt Synthesis for Large Language Model Reasoning
-
Anthropic Releases Claude Sonnet 4.5, World’s Best Coding Model
By
–
BREAKING: Anthropic just released Claude Sonnet 4.5, calling it the world’s best coding model • Achieved 82% accuracy on SWE-bench Verified, topping GPT-5 and Gemini 2.5 Pro • Strongest yet at building complex AI agents and using real computers for tasks • Claude Code
-
DoRA Optimization Technique Shows Better Results Potential
By
–
IIUC they didn't use DoRA, so the results can be even better than shown here:
-
Weight decay ineffective for value-based reinforcement learning
By
–
No, I haven’t seen any reliable wins from weight decay for value based RL, and I haven’t looked several times.