algorithm

Proposes Efficient Long-horizon Optimization (ELO) learning, a meta-training algorithm that reallocates compute to longer horizons and uses decoupled progressive expert supervision, improving learned optimizers' performance on long-unroll tasks and out-of-distribution generalization. ELO-Celo2 consistently outperforms AdamW and matches Muon on language modeling tasks.

A new study finds that while TikTok's 'not interested' button effectively reduces unwanted content by 84%, the algorithm can slowly reintroduce such content if users do not maintain consistent feedback. Researchers cloned accounts to study personalization, highlighting limits on user agency.

This thread discusses modifications to GRPO for agentic RL, focusing on different levels of advantage normalization (prompt-level, task-level, environment-level) to handle higher reward variance in multi-task, multi-turn environments.

An X platform executive addresses claims that X is becoming like TikTok, stating that video posts already drive nearly half of impressions, many videos are recycled, and a new video editor is being introduced to encourage original content.

The article describes an expired patent for a new in-memory sort algorithm that features common prefix skipping, adaptivity, and key substring caching, which was implemented in Oracle 10gR2 and achieved significant performance improvements.

Discusses a 2014 paper that refutes the 3SUM conjecture by presenting subquadratic algorithms for the 3SUM problem, with implications for computational geometry and graph algorithms.

Explains how to use the A* algorithm for pathfinding around circular obstacles by converting the environment into a graph using tangent visibility and bitangent lines.

This paper presents a conflict-free path-planning algorithm for en-route air traffic control, designed to be interpretable and computationally efficient for human operators. The algorithm integrates three conflict detection methods and achieves fast computation times, demonstrated on a real-world sector.

This paper presents an algorithm for group distributionally robust least squares regression using block Lewis weights, achieving improved complexity over interior point methods. It also provides interpolating algorithms between average and robust losses.

This paper presents a polynomial-time algorithm for learning the structure of a Gaussian graphical model from a single trajectory of Glauber dynamics, with a trajectory-length guarantee that does not depend on the mixing time.

This article revisits Joel Yliluoma's 2011 ordered dither algorithm, explains its internals in detail, presents simplified variants, and compares results to state-of-the-art. Source code is included.

A Python implementation of the Gumbel-Sinkhorn neural network for sorting lists of numbers, based on the 2018 paper by Mena et al.

The Chinese edition of Introduction to Algorithms (4th Edition) is finally published, split into two volumes: Core Chapters and Extended Chapters, corresponding to Chapters 1–25 and Chapters 26–35 of the original book.

Explores parallel algorithms for matching parentheses, a fundamental problem in compilers and text processing.

This paper proposes a resource-light algorithm to automatically assign part-of-speech tags to senses in the Al-Mawrid Arabic-English bilingual dictionary by transferring tags from English WordNet after disambiguation, achieving high accuracy with minimal cost.

User BenjaminDEKR notes that despite having 43,600 followers (36% verified), X showed his most recent post to only 780 people, questioning the platform's algorithm.

A 2013 video visualizing and audibilizing 15 sorting algorithms in 6 minutes, including selection sort, quick sort, and bogo sort.

RQL is a new algorithm for offline flow reinforcement learning that treats flow steps as MDP steps and uses reversed flows to generate hindsight trajectories.

Threads has surpassed 500 million monthly active users, and Meta is introducing new features like Communities and 'Your Algo' to give users more control over their algorithm.

Elon Musk shares his 5-step algorithm for engineering problem-solving, emphasizing questioning requirements, deleting unnecessary steps, then optimizing, speeding up, and automating.