Logo des Repositoriums

Auflistung nach Schlagwort "data-centric AI"

1 - 2 von 2
  • Konferenzbeitrag
    ATDLLMD: Acceptance test-driven LLM development
    (Softwaretechnik-Trends Band 44, Heft 2, 2024) Faragó, David
    Since the capabilities of Large Language Models (LLMs) have massively increased in the last years, many new applications based on LLMs are possible. However, these new applications also pose new challenges in LLM development. This article proposes an acceptance test-driven development (ATDD) style, baptized ATDLLMD, where the LLM’s training and test sets are extended in each iteration by data coming from validation of the previous iteration’s LLM and system around the LLM. So the validation phase supplies the additional or updated data for training and verification of the LLM. ATDLLMD is made possible by two major innovative solutions: applying the innovative CPMAI process, and applying our own verification tool, LM-Eval, leading to a red-train green cycle for LLM development, which resembles ATDD, but integrates data science best practices.
  • Konferenzbeitrag
    A High Quality Data Pipeline for Reasonable-Scale Machine Learning
    (Softwaretechnik-Trends Band 42, Heft 4, 2022) Faragó, David
    Data quality (especially correctness) plays a critical role in the success of a machine learning (ML) project. This paper describes a data pipeline for creating high quality data, using as example Key Information Extraction (KIE) from invoices – one of the most popular tasks in Intelligent Document Processing (IDP). The tasks of each data pipeline step are listed, showing the decisions and technology involved. The focus is on practicality: doing ML at reasonable-scale, i.e. with as little cost (people and hardware) as possible, and a concern for practice more than achieving high scores on a metric that is not grounded in practical use. Contributions: 1. an extended list of quality dimensions, with simple definitions 2. overview of a data pipeline, examplified on KIE 3. for each pipeline step a list of tasks, showing decisions, pitfalls, and technology involved 4. in particular, how to use the state of the art contrastive model CLIP to solve difficult selection and reduction tasks on images 5. a tool for labeling key information on images 6. a labeling guide for invoices. Most contributions can easily be transfered to other supervised learning tasks.