ML Notes — English https://notes.iwase.dev/en/ Machine learning and deep learning research notes (English edition) quarto-1.10.18 Sat, 08 Aug 2026 00:00:00 GMT Harness Optimization Naoto Iwase https://notes.iwase.dev/en/harness-optimization/ An LLM agent’s behavior is not determined when its model is selected. What enters context, how tools are provided, when outputs are verified, and how the agent recovers from failure all matter. Even with the same model, the surrounding execution system changes which tasks it can solve, its cost, and its failure modes. This book calls that execution system, comprising prompts, memory, control flow, validators, and recovery logic, an executable harness.

Can the harness itself be improved from execution records and evaluations without changing the model? This book treats edits to prompts, workflows, memory, and runtime code as a single optimization loop. It generates candidate updates from failures, identifies real improvements under limited budgets and noisy evaluations, and carries only verified changes forward to the next execution. The goal is not for the agent merely to change. The goal is to make that change a reproducible, reliable improvement.

]]> LLM AI Agent https://notes.iwase.dev/en/harness-optimization/ Sat, 08 Aug 2026 00:00:00 GMT Small Recursive Reasoning Models Naoto Iwase https://notes.iwase.dev/en/recursive-reasoning/ A distinct line of reasoning models uses small neural networks that are recursively unrolled at test time. These models seek to trade parameter count for iterative computation and have been studied on structured tasks such as Sudoku and ARC-AGI.

This book examines recursive reasoning models through their technical foundations, prior art, empirical evaluation, and critiques. It also connects them to latent reasoning, test-time scaling, implementation choices, and open problems.

]]> Deep Learning Reasoning https://notes.iwase.dev/en/recursive-reasoning/ Sun, 24 May 2026 00:00:00 GMT Reliable Reasoning Naoto Iwase https://notes.iwase.dev/en/reliable-reasoning/ Research on making Large Language Model (LLM) reasoning more reliable expanded rapidly in 2025–2026. This book connects methods that use external signals to improve or evaluate reasoning rather than treating longer generation alone as evidence of progress.

It organizes the literature around training-side signals, inference-side signals, and structural approaches, with particular attention to capability gains, correctness estimation, and the allocation of inference compute.

]]> LLM Reasoning https://notes.iwase.dev/en/reliable-reasoning/ Tue, 19 May 2026 00:00:00 GMT Diffusion Language Models Naoto Iwase https://notes.iwase.dev/en/dllm/ Diffusion Language Models (DLLM) bring the ideas behind the diffusion models that succeeded in image generation into language modeling. Recent years have seen large-scale implementations such as LLaDA and Dream, alongside commercial-grade systems including Mercury and Gemini Diffusion. This book consolidates the key references needed to understand modern DLLMs, integrating the taxonomy presented in the Li et al. 2025 survey, and systematically covers formulation, sampling, the correspondence with continuous diffusion, adaptation from AR models, derivative discrete models, hybrid architectures, inference acceleration, guidance, post-training, multimodal extensions, and downstream applications.

]]> LLM Generative Model https://notes.iwase.dev/en/dllm/ Fri, 15 May 2026 00:00:00 GMT One-Step Generation Naoto Iwase https://notes.iwase.dev/en/one-step-generation/ Between 2025 and 2026, methods that overcome the multi-step inference of diffusion models and Flow Matching to generate high-quality images with a single network evaluation (1-NFE) have been rapidly advancing. This series curates four papers driving this field, tracing the technical evolution from extensions of Flow Matching to entirely new paradigms.

]]> Deep Learning Generative Model https://notes.iwase.dev/en/one-step-generation/ Wed, 11 Feb 2026 00:00:00 GMT Molmo2 Naoto Iwase https://notes.iwase.dev/en/molmo2/ Molmo2 (Multimodal Open Language Model 2) is a fully open Vision-Language Model (VLM) family developed by the Allen Institute for AI and the University of Washington. Its defining capability is video grounding: identifying when and where an event or object appears within a video.

This book examines the open data pipeline, model family, and grounding evaluations behind that capability, including the reported comparisons with proprietary systems.

Paper · Code · Demo

]]> VLM https://notes.iwase.dev/en/molmo2/ Tue, 03 Feb 2026 00:00:00 GMT Olmo 3 Naoto Iwase https://notes.iwase.dev/en/olmo-3/ Olmo 3 is a family of state-of-the-art, fully-open language models at the 7B and 32B parameter scales developed by the Allen Institute for AI (AI2). This release includes the entire Model Flow, i.e., the full lifecycle of the family of models, including every stage, checkpoint, data point, and dependency used to build it.

Paper: arXiv:2512.13961

]]> LLM Reasoning https://notes.iwase.dev/en/olmo-3/ Mon, 02 Feb 2026 00:00:00 GMT