Remote AI Paper Discussion Group

Link: https://arxiv.org/abs/2608.27454
Title: WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution
Track: Self-Improving LLM Agents
Content: 1) WikiSkill improves agent-skill evolution by maintaining a persistent “wiki” that consolidates lessons from past executions, separating raw experience, accumulated knowledge, and executable skills so later skill updates can systematically reuse earlier insights. Across benchmarks, this persistent knowledge accumulation improves skill evolution, transfers well across models and model families, and can let smaller models with evolved skills outperform much larger models without skills. 2) Reflection on Track: Resummarize and compare all 7 papers we discussed in this self-improving LLM agents track.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.

In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.

Link: https://arxiv.org/pdf/2506.14202
Title: DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretation
Track: Sakana
Content: DiffusionBlocks turns transformer networks into independently trainable blocks by interpreting residual updates as a denoising process and training each block with a score-matching objective. This enables one-block-at-a-time gradient computation, reducing memory roughly in proportion to the number of blocks while matching end-to-end training performance across diverse transformer architectures and generative tasks.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.

In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.

Link: https://arxiv.org/pdf/2601.03335
Title: Digital Red Queen:Adversarial Program Evolution in Core War with LLMs
Track: Sakana
Content: Digital Red Queen (DRQ) uses LLMs to evolve Redcode “warriors” that compete in Core War, a self-modifying virtual-machine environment where each program tries to crash its opponents and remain the last one running. Across successive training rounds, new warriors compete against all earlier ones, and independent DRQ runs tend to converge on similar behaviors while producing increasingly robust strategies.
Slack link: ml-ka.slack.com, channel: #pdg. Please join us -- if you cannot join, please message us here or to mlpaperdiscussiongroupka@gmail.com.

In the Paper Discussion Group (PDG) we discuss recent and fundamental papers in the area of machine learning on a weekly basis. If you are interested, please read the paper beforehand and join us for the discussion. If you have not fully understood the paper, you can still participate – everyone is welcome! You can join the discussion or simply listen in. The discussion is in German or English depending on the participants.