PGT projects 2024

Theme 1: Data-centric AI explainability

Context. Research around Explainable AI (XAI) is interpreted differently depending on the type of model, data, and applications. In general, however, it has been focused primarily on explaining model inference (see eg LIME, occlusion testing for images), with relatively little relevance given to linking the inference to the data used for training.

This is changing, with recent advances in the area of so-called "Data-Centric AI". For example, concepts underpinning data valuation, such as Influence Functions [WFW+20] and more recently AME (Average Marginal Effects) [LLZ+22] help pinpoint specific data points in the training set, which are most responsible for a given inference. Other, relatively older concepts like Explanation Tables [EF+18] have also been "resurrected" with the aim to provide data-centric explanations.

References

• [WFW+20] Wu, Weiyuan, Lampros Flokas, Eugene Wu, and Jiannan Wang. ‘Complaint-Driven Training Data Debugging for Query 2.0’. In Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, 1317–34, 2020. https://dl.acm.org/doi/abs/10.1145/3318464.3389696.

• [LLZ+22] Lin, Jinkun, Anqi Zhang, Mathias Lécuyer, Jinyang Li, Aurojit Panda, and Siddhartha Sen. ‘Measuring the Effect of Training Data on Deep Learning Predictions via Randomized Experiments’. In Proceedings of the 39th International Conference on Machine Learning, 13468–504. PMLR, 2022. https://proceedings.mlr.press/v162/lin22h.html.

• [EF+18] El Gebaly, Kareem, Guoyao Feng, Lukasz Golab, Flip Korn, and Divesh Srivastava. ‘Explanation Tables’. Sat 5 (2018): 14.

Starting point:

We have been developing a tool for capturing the derivations of data through python/Pandas scripts: https://github.com/Lucass97/data_provenance_for_data_science.

The tool is described in this recent presentation: https://www.slideshare.net/slideshow/design-and-development-of-a-provenance-capture-platform-for-data-science/268330702

Concrete projects:

Using the above as a starting point, we can develop ideas in a number of interesting directions.  Please see here for a recent talk that provides more background: https://www.slideshare.net/slideshow/explainable-data-centric-ai-what-are-you-explaininhg-and-to-whom/268463441

• LLMs exploiting the derivation graphs to provide suggestions on pipeline and data repairs.  This project will look at using RAG (Retrieval Augmented Generation), specifically to incorporate results from Neo4J queries (Cypher) into narratives that suggest interventions on data and on pipelines

• Graph analysis of provenance graphs.  The derivation graph is natively stored in a Neo4J graph DB. The project will experiment with graph analysis algorithms on these derivations, using the Neo4J Graph Data Science library

• "Why+" explanations: augmenting data derivations to describe the behaviour of complex data processing algorithms, for example training set optimisation, incremental data cleaning, etc. please see this presentation paper: https://www.dropbox.com/scl/fi/yfpzxtsbrtj9oppc52ymk/DCAI_position_SEBD_24_CR.pdf

I am open to discussing other related ideas for alternative projects.