Details

Reševanje nalog iz korpusa ARC-AGI-1 z metodami umetne inteligence
ID Praček, Marko (Author), ID Žabkar, Jure (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (1,14 MB)
MD5: A72393E8036FA9EA750851909F261196

Abstract
V diplomski nalogi obravnavamo reševanje nalog korpusa ARC-AGI-1 (Abstraction and Reasoning Corpus), ki je merilo zmogljivosti za abstraktno razmišljanje in sklepanje umetne inteligence iz malo primerov. Predstavimo hibridni sistem, v katerem veliki jezikovni model (LLM) s pomočjo DSPy ustvari pogoje in akcije, jih izvaja nad matrikami ter jih iterativno popravlja in razširja na podlagi poročila razlik. Namesto enumeracije v DSL uporabimo izboljševalno zanko, ki s pomočjo preiskovanja prostora navodil išče rešitve za specifične naloge. Na vzorcu nalog iz ARC-AGI-1 smo z modelom GPT-5-mini dosegeli 61,5% točnost na 60 učnih nalogah in 43% na 100 javnih evalvacijskih nalogah.

Language:Slovenian
Keywords:umetna inteligenca, Abstraction and Reasoning Corpus, učenje na malo učnih primerih, veliki jezikovni modeli
Work type:Bachelor thesis/paper
Typology:2.11 - Undergraduate Thesis
Organization:FRI - Faculty of Computer and Information Science
Year:2025
PID:20.500.12556/RUL-173254 This link opens in a new window
COBISS.SI-ID:250405635 This link opens in a new window
Publication date in RUL:15.09.2025
Views:606
Downloads:185
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:English
Title:Solving Tasks from the ARC-AGI-1 Corpus Using Artificial Intelligence
Abstract:
In this thesis, we address solving ARC (Abstraction and Reasoning Corpus) tasks, which measure abstract reasoning from few examples. We present a hybrid system in which a large language model (LLM), assisted by DSPy, generates preconditions and actions, executes them on matrices, and iteratively repairs and adds them based on a difference report. Instead of enumerating programs in a DSL, we use an improvement loop that explores the space of instructions to find solutions for specific ARC tasks. On a sample of tasks from ARC-AGI-1, the GPT-5-mini model achieved 61.5% accuracy on 60 training tasks and 43% on 100 public evaluation tasks.

Keywords:artificial intelligence, Abstraction and Reasoning Corpus, few-shot learning, large language models

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back