Alaia Solko-Breslin

Stop by on Thursday evening poster session at #NeurIPS2024 to learn about our work on how to integrate black box components in learning pipelines

Happy to present our recent work: "Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference" 📝 Blog: debugml.github.io/logicbreaks/ 🧐 arXiv: arxiv.org/abs/2407.00075 🤖 Code: github.com/AntonXue/tf_lo…

Excited to present our ICML 2024 paper: "Towards Compositionality in Concept Learning"! 🔗 Blog: debugml.github.io/compositional-… 📄 Paper: arxiv.org/abs/2406.18534 💻 Code: github.com/adaminsky/comp…

Joint work with Seewon Choi, @_ziyang_ , @NeelayV , @RajeevAlur , @AI4Code , @RICEric22.

We develop ISED, an algorithm for data-efficient learning of neural programs without intermediate labels. ISED relies only on input-output samples to estimate the gradient of the program, making it more general than neurosymbolic frameworks that are restricted to logic programs.

Introducing neural programming: end-to-end learning of neural models composed with any black-box program, even those that call GPT-4. Paper: arxiv.org/abs/2406.06246 Blog: debugml.github.io/neural-program… Code: github.com/alaiasolkobres…