Hallucination Detection in LLMs Using Spectral Features of Attention Maps

1 code implementation • 24 Feb 2025 • Jakub Binkowski, Denis Janiak, Albert Sawczyn, Bogdan Gabrys, Tomasz Kajdanowicz

We propose the $\text{LapEigvals}$ method, which utilises the top-$k$ eigenvalues of the Laplacian matrix derived from the attention maps as an input to hallucination detection probes.

Hallucination

Empowering Small-Scale Knowledge Graphs: A Strategy of Leveraging General-Purpose Knowledge Graphs for Enriched Embeddings

no code implementations • 17 May 2024 • Albert Sawczyn, Jakub Binkowski, Piotr Bielak, Tomasz Kajdanowicz

Knowledge-intensive tasks pose a significant challenge for Machine Learning (ML) techniques.

Entity Alignment Representation Learning

Unveiling the Potential of Probabilistic Embeddings in Self-Supervised Learning

no code implementations • 27 Oct 2023 • Denis Janiak, Jakub Binkowski, Piotr Bielak, Tomasz Kajdanowicz

In recent years, self-supervised learning has played a pivotal role in advancing machine learning by allowing models to acquire meaningful representations from unlabeled data.

Out-of-Distribution Detection Self-Supervised Learning

Graph-level representations using ensemble-based readout functions

1 code implementation • 3 Mar 2023 • Jakub Binkowski, Albert Sawczyn, Denis Janiak, Piotr Bielak, Tomasz Kajdanowicz

Graph machine learning models have been successfully deployed in a variety of application areas.

RAFEN -- Regularized Alignment Framework for Embeddings of Nodes

1 code implementation • 3 Mar 2023 • Kamil Tagowski, Piotr Bielak, Jakub Binkowski, Tomasz Kajdanowicz

A well-defined node embedding model should reflect both node features and the graph structure in the final embedding.