Is it possible to train a DQN to solve a maze with non-convex obstacles in a long-horizon navigation task (in 10 minutes or less)? The rules are: You can not use old data except for the replay buffer The inputs are only the x and y coordinates of the state and the distance of the agent to the goal Step size should not exceed 2% of the total maze size You must start from the same initial state […]
arXiv:2503.00229v4 Announce Type: replace-cross Abstract: Armijo line-search (Armijo-LS) is a standard method to set the step-size for gradient descent (GD). For smooth functions, Armijo-LS alleviates the need to know the global smoothness constant L and adapts to the “local” smoothness, enabling GD to converge faster. Existing theoretical analyses show that GD with Armijo-LS (GD-LS) can result in constant factor improvements over GD with a 1/L step-size (denoted as GD(1/L)). We strengthen these results and show that if the […]
Classical approximation bases such as Chebyshev polynomials provide principled and interpretable representations, but their multivariate tensor-product constructions scale exponentially with dimension and impose axis-aligned structure that is poorly matched to real tabular data. We address this by replacing tensorised oscillations with directional harmonic modes of the form $cos(mathbf{m}^{top}arccos(mathbf{x}))$, which organise multivariate structure by direction in angular space rather than by coordinate index. This representation yields a discrete spectral regression model in which complexity is controlled by selecting a […]
In this edition of our Haskell in Production series, we sit down with RELEX Solutions, a supply chain & retail planning platform. In this interview with Serokell, Lead Software Developers, Mats Rauhala and Martin Potier, shared with us their reasons for choosing Haskell for their project and their experience implementing it. Interview with RELEX Solutions Could you give our readers a brief introduction to RELEX Solutions and your role there? Martin: RELEX Solutions edits a fairly complete supply […]
As Large Language Models (LLMs) achieve remarkable empirical success through scaling model and data size, pretraining has become increasingly critical yet computationally prohibitive, hindering rapid development. Despite the availability of numerous pretrained LLMs developed at significant computational expense, a fundamental real-world question remains underexplored: textit{Can we leverage existing small pretrained models to accelerate the training of larger models?} In this paper, we propose a Late-to-Early Training (LET) paradigm that enables LLMs to explicitly learn later knowledge in earlier […]
arXiv:2601.10881v1 Announce Type: new Abstract: In the spanning-tree congestion problem ($mathsf{STC}$), we are given a graph $G$, and the objective is to compute a spanning tree of $G$ that minimizes the maximum edge congestion. While $mathsf{STC}$ is known to be $mathbb{NP}$-hard, even for some restricted graph classes, several key questions regarding its computational complexity remain open, and we address some of these in our paper. (i) For graphs of degree at most $d$, it is known that $mathsf{STC}$ […]
Hey everyone, I’ve been working on an RL project where I trained a PPO agent to control 4 elevators in a 20-floor building simulation. The goal was to see if RL can beat a classical Destination Dispatching algorithm. Results after 5M training steps on CPU: Classic agent: mean reward -0.67, avg wait 601 steps PPO agent: mean reward +0.14, avg wait 93 steps (~84% reduction) The hardest part was reward engineering – took several iterations to get dense […]
How Summaries, KPIs, and Portfolio Metrics Are Constructed and Distorted Every business decision starts with a question. What are our total sales by region? Which product categories generate the most revenue? How do customer segments compare in profitability? These questions all share something in common: they require grouping data and calculating aggregates. Grouping and aggregation are the backbone of business intelligence. They transform raw transactions into actionable insights. They turn thousands of individual data points into clear summaries […]
The weakest link in the privacy of cryptocurrency transactions is often outside the blockchain. There are technologies such as stealth addresses and subaddresses to try to thwart attempts to link transactions to individuals. They do a good job of anonymizing transaction data, but the weak link may be metadata, as is often the case. Cryptocurrency nodes circulate transaction data using a peer-to-peer network. An entity running multiple nodes can compare when data arrived at each of its nodes […]
Scientists have discovered that male fireflies in a South Carolina swamp follow local interaction rules to synchronize their flashing mating displays. The research is being presented at a meeting of the American Physical Society in Denver. (A preprint is also available on the biorxiv.) Such work could one day lead to insights into how the body’s cells sync to its internal circadian rhythm, or how neurons fire together in the brain, as well as the design of drone […]