arXiv subjects

Ekaterina Trofimova

Publications and source records attributed to Ekaterina Trofimova.

7 recordsLinked to original sources

arXiv2026

Citations Are Late: Reading epistemic instability from what papers believe, years before the citation graph catches up

Paradigm shifts in science are visible in what researchers assert and contest before they are visible in the citation graph. We ask whether a cheap, content-level signal of epistemic instability, derived from the changing distribution of stated modelling beliefs in paper abstracts, can anticipate a paradigm shift earlier than the dominant citation-based disruption index (CD5). On the displacement of recurrent networks by Transformers in NLP, a pre-registered content signal crosses its detection threshold in 2016-Q1, whereas a real-time CD5 monitor cannot even observe the 2017 breakthrough until 2022-Q2, since CD5 needs a five-year forward-citation window: a lead of about 25 quarters. The flat citation baseline is not an artifact of one index, as our CD5, a reference-normalised variant, and an authoritative precomputed index all sit near zero across the shift. The lead is also not a faster proxy: at equal latency the signal beats the content competitors tested, including a learned CD-from-text model and an embedding disruption measure, and the decomposition sees what a keyword cannot (keyword-blind AUC of about 0.9, reproduced on human labels). The lead over CD5 is an observability lead. Which signal carries it depends on the shift: belief adoption in NLP, contestation and applicability stress in vision. Measured against the breakthroughs themselves, the signal leads by five quarters in NLP and is contemporaneous in computer vision.

cs.DL↗

arXiv2026

ML2B: Benchmarking LLMs on Cross-Lingual ML Pipeline Generation

We introduce ML2B, the first benchmark for evaluating cross-lingual task comprehension in end-to-end ML pipeline generation by large language models. Despite growing global AI adoption, no systematic evaluation exists for ML pipeline generation beyond English task descriptions. ML2B addresses this gap with 35 Kaggle competitions spanning tabular, text, and image domains, translated into 14 languages by native-speaker researchers with ML expertise, yielding 490 task-language pairs. To ensure evaluation integrity, the benchmark incorporates 10 private competitions without publicly available solutions and employs network-isolated evaluation infrastructure restricting runtime access to essential ML resources. We provide standardized evaluation protocols, an AutoGluon algorithmic baseline, and comprehensive failure mode analysis. Experiments with frontier models (GPT-4.1-mini, GPT-OSS-120b, Gemini-2.5-Flash) reveal that cross-lingual performance degradation is highly task-dependent rather than following traditional resource-availability hierarchies, with gaps ranging from language advantages to severe degradation depending on competition characteristics. These findings challenge conventional assumptions about multilingual model capabilities and underscore the necessity of systematic cross-lingual evaluation for ML pipeline generation. We open-source the benchmark, baselines, and evaluation infrastructure at https://github.com/enaix/ml2b.

cs.CL↗

arXiv2026

CodeRefine: A Pipeline for Enhancing LLM-Generated Code Implementations of Research Papers

This paper presents CodeRefine, a novel framework for automatically transforming research paper methodologies into functional code using Large Language Models (LLMs). Our multi-step approach first extracts and summarizes key text chunks from papers, analyzes their code relevance, and creates a knowledge graph using a predefined ontology. Code is then generated from this structured representation and enhanced through a proposed retrospective retrieval-augmented generation approach. CodeRefine addresses the challenge of bridging theoretical research and practical implementation, offering a more accurate alternative to LLM zero-shot prompting. Evaluations on diverse scientific papers demonstrate CodeRefine's ability to improve code implementation from the paper, potentially accelerating the adoption of cutting-edge algorithms in real-world applications.

cs.CL↗

arXiv2025

TGPR: Tree-Guided Policy Refinement for Robust Self-Debugging of LLMs

Iterative refinement has been a promising paradigm to enable large language models (LLMs) to resolve difficult reasoning and problem-solving tasks. One of the key challenges, however, is how to effectively search through the enormous search space of possible refinements. Existing methods typically fall back on predefined heuristics, which are troubled by the exploration-exploitation dilemma and cannot adapt based on past refinement outcomes. We introduce Tree-Guided Policy Refinement (TGPR), a novel framework that combines GRPO with a Thompson-Sampling-based tree search. TGPR explores both failed and successful refinement paths actively, with denser training trajectories and more adaptive policies. On HumanEval, MBPP, and APPS benchmarks, our method achieves up to +4.2 percentage points absolute improvement in pass@1 (on MBPP) and up to +12.51 percentage points absolute improvement in pass@10 (on APPS) compared to a competitive GRPO baseline. Apart from debugging code, TGPR focuses on a principled approach to combining learned policies with structured search methods, offering a general framework for enhancing iterative refinement and stateful reasoning in LLMs.

cs.AI↗

arXiv2024

Linguacodus: A Synergistic Framework for Transformative Code Generation in Machine Learning Pipelines

In the ever-evolving landscape of machine learning, seamless translation of natural language descriptions into executable code remains a formidable challenge. This paper introduces Linguacodus, an innovative framework designed to tackle this challenge by deploying a dynamic pipeline that iteratively transforms natural language task descriptions into code through high-level data-shaping instructions. The core of Linguacodus is a fine-tuned large language model (LLM), empowered to evaluate diverse solutions for various problems and select the most fitting one for a given task. This paper details the fine-tuning process, and sheds light on how natural language descriptions can be translated into functional code. Linguacodus represents a substantial leap towards automated code generation, effectively bridging the gap between task descriptions and executable code. It holds great promise for advancing machine learning applications across diverse domains. Additionally, we propose an algorithm capable of transforming a natural description of an ML task into code with minimal human interaction. In extensive experiments on a vast machine learning code dataset originating from Kaggle, we showcase the effectiveness of Linguacodus. The investigations highlight its potential applications across diverse domains, emphasizing its impact on applied machine learning in various scientific fields.

cs.LG↗

arXiv2022

Code4ML: a Large-scale Dataset of annotated Machine Learning Code

Program code as a data source is gaining popularity in the data science community. Possible applications for models trained on such assets range from classification for data dimensionality reduction to automatic code generation. However, without annotation number of methods that could be applied is somewhat limited. To address the lack of annotated datasets, we present the Code4ML corpus. It contains code snippets, task summaries, competitions and dataset descriptions publicly available from Kaggle - the leading platform for hosting data science competitions. The corpus consists of ~2.5 million snippets of ML code collected from ~100 thousand Jupyter notebooks. A representative fraction of the snippets is annotated by human assessors through a user-friendly interface specially designed for that purpose. Code4ML dataset can potentially help address a number of software engineering or data science challenges through a data-driven approach. For example, it can be helpful for semantic code classification, code auto-completion, and code generation for an ML task specified in natural language.

cs.SE↗

arXiv2021

Segmentation of EM showers for neutrino experiments with deep graph neural networks

We introduce a first-ever algorithm for the reconstruction of multiple showers from the data collected with electromagnetic (EM) sampling calorimeters. Such detectors are widely used in High Energy Physics to measure the energy and kinematics of in-going particles. In this work, we consider the case when many electrons pass through an Emulsion Cloud Chamber (ECC) brick, initiating electron-induced electromagnetic showers, which can be the case with long exposure times or large input particle flux. For example, SHiP experiment is planning to use emulsion detectors for dark matter search and neutrino physics investigation. The expected full flux of SHiP experiment is about 10^20 particles over five years. To reduce the cost of the experiment associated with the replacement of the ECC brick and off-line data taking (emulsion scanning), it is decided to increase exposure time. Thus, we expect to observe a lot of overlapping showers, which turn EM showers reconstruction into a challenging point cloud segmentation problem. Our reconstruction pipeline consists of a Graph Neural Network that predicts an adjacency matrix and a clustering algorithm. We propose a new layer type (EmulsionConv) that takes into account geometrical properties of shower development in ECC brick. For the clustering of overlapping showers, we use a modified hierarchical density-based clustering algorithm. Our method does not use any prior information about the incoming particles and identifies up to 87% of electromagnetic showers in emulsion detectors. The main test bench for the algorithm for reconstructing electromagnetic showers is going to be SND@LHC.

cs.LG↗