Daily report — 2026-10-02

As an alternative to the standard geometric analyses, we give an exact, information-theoretic analysis of stochastic gradient descent (SGD) and its variants. We show that a preconditioned SGD step is the posterior-mean update of a Gaussian Bayes model, and that its one-step regret splits into an intrinsic-time cost and a change in comparator information.

Claimed: none extracted

Evidence: + cheap/closed-form ("Convex"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed-form"); + formal statement ("convergence"); no public code found; + keywords: optimizer

heuristic

Several state-of-the-art methods for online reinforcement learning in continuous control improve policies using action gradients of a learned critic. However, critics are typically trained to predict returns, and accurate value predictions do not necessarily yield accurate action derivatives, potentially leading to unreliable policy updates.

Claimed: none extracted

Evidence: + public benchmarks ("benchmarks"); + explicit algorithm/pseudocode ("Algorithm 1"); + formal statement ("theorem"); + code: https://github.com/Atarilab/FERPO; + keywords: sampling, regularization

heuristic

Counterfactual generation seeks to sample outcomes under a hypothetical intervention or decision using observational data collected under the factual assignment mechanism. We develop a flow-matching approach that combines a sample-split, doubly robust training objective with a learned coupling between observed source outcomes and target outcomes drawn from a fitted conditional outcome model.

Claimed: none extracted

Evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + explicit algorithm/pseudocode ("Algorithm 2"); + formal statement ("bound"); no public code found; + keywords: monte carlo

heuristic

Discrete diffusion language models offer a compelling alternative to autoregressive generation for tasks demanding bidirectional reasoning and global constraint satisfaction. Yet they share a structural bottleneck: when decoding in parallel, each token is sampled independently from its marginal, severing the statistical dependencies among the tokens decoded together.

Claimed: perplexity 1.0

Evidence: - diffusion/video ("Diffusion"); + concrete method ("closed-form"); + formal statement ("bound"); + claims simplicity ("simple"); + code: https://github.com/rhfeiyang/HC-DLM; + keywords: variational; + 50 HF upvotes

heuristic

Decision-focused learning for linear optimization is complicated by the discontinuity of the optimizer, where small cost errors may leave the decision unchanged or move it to a different vertex. We show that this non-smooth pointwise behavior becomes locally quadratic after averaging over the data distribution, and we derive the curvature in closed form, specifically, a matrix-valued measure suppo

Claimed: none extracted

Evidence: + cheap/closed-form ("closed form"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed form"); + formal statement ("We prove"); + claims simplicity ("simple"); no public code found; + keywords: optimizer

heuristic

On-policy self-distillation has recently emerged as an effective approach for improving language-model reasoning by supervising students with a frozen or EMA version of themselves that receives privileged information. Its application to multimodal large language models (MLLMs), however, remains largely unexplored.

Claimed: none extracted

Evidence: - LLM-based ("large language model"); + toy/synthetic experiments ("Synthetic"); + small/synthetic data ("Synthetic"); + public benchmarks ("benchmarks"); + claims simplicity ("simple"); + code: https://github.com/sirkosophia/Where-OPD

heuristic

Masked diffusion models (MDMs) generate sequences by progressively unmasking several tokens per denoising step, but their reverse process is typically factorized over positions, limiting sample quality in the few-step regime where diffusion's speed advantage over autoregressive decoding matters most. A recent line of work introduces a continuous Gaussian latent, trained as a variational autoencode

Claimed: none extracted

Evidence: - diffusion/video ("Diffusion"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("bound"); no public code found; + keywords: variational; + 17 HF upvotes

heuristic

Step-size selection remains a central challenge in large-scale neural network optimization; conservative steps slow convergence, while aggressive steps can destabilize it. We combine \textbf{Z}ero-and-\textbf{F}irst-\textbf{O}rder optimization~(ZFO) and propose a lightweight framework that decouples direction selection from step-size.

Claimed: none extracted

Evidence: - LLM-based ("LLM"); + CPU-friendly ("lightweight"); + explicit algorithm/pseudocode ("Algorithm 1"); + formal statement ("convergence"); - system/framework (many moving parts) ("framework"); + code: https://github.com/nizswan/Zeroth-First-Order-Framework; + keywords: optimizer

heuristic

Intrinsic rewards are designed to guide exploration in reinforcement learning by assigning value to an agent's experience, for example through prediction error or learning progress. However, maximizing these rewards need not produce the most informative experience available.

Claimed: none extracted

Evidence: + formal statement ("bound"); + claims simplicity ("simple"); + code: https://github.com/scottviteri/what-is-exploration

heuristic

Non-Markovian environments are often modeled as Regular Decision Processes (RDPs), where dynamics depend on the interaction history through a finite automaton. Existing offline guarantees for RDPs rely on a distinguishability assumption on the behaviour policy but provide no means of verifying it.

Claimed: none extracted

Evidence: + formal statement ("We prove"); + code: https://github.com/Kcbir/pec

heuristic