Logical Constraints as Inductive Biases in Learning and Generation | Nijesh Upreti

💬 Discord: https://discord.gg/h8NVzwnysW 💻 GitHub: https://github.com/centaurinstitute 🤝 LinkedIn: https://www.linkedin.com/company/centaur-ai-institute/ 📢 New Initiative: Neuro-Symbolic Agentic Protocol ⭐ Star us on GitHub https://github.com/centaurinstitute/neuro-symbolic-agentic-protocol — ​​​​​Abstract: Neuro-symbolic AI seeks to combine the statistical strengths of machine learning with the structure and interpretability of symbolic reasoning. In this talk, I will present two recent projects that explore this interaction from complementary perspectives. The first project develops a semantic framework for weak supervision based on Inductive Logic Programming (ILP). Focusing on multi-instance partial label learning, the work formalizes how logical constraints can structure ambiguous supervision, define hypothesis spaces over latent label transitions, and provide a framework for reasoning about consistency and failure modes in weakly supervised learning. The second project explores logic-guided generative modeling through Logic Tensor Network-Enhanced GANs (LTN-GANs). By incorporating differentiable logical constraints into adversarial training, the framework guides generation toward outputs that satisfy domain-specific rules while maintaining sample diversity and quality. Taken together, these projects illustrate a common theme: symbolic structure can provide useful inductive biases when supervision is incomplete, ambiguous, or underspecified. I conclude by discussing broader implications for neuro-symbolic learning and future directions toward more reliable and interpretable AI systems. Paper Links: Upreti, N., & Belle, V. Neuro-symbolic weak supervision: Theory and semantics. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences. https://arxiv.org/abs/2503.18509 Upreti, N., & Belle, V. Logic tensor network-enhanced generative adversarial network. In The 41st International Conference on Logic Programming. (Electronic Proceedings in Theoretical Computer Science). Open Publishing Association. https://arxiv.org/abs/2601.03839 Bio: Nijesh Upreti is a doctoral researcher at the University of Edinburgh working in neuro-symbolic AI under the supervision of Professor Vaishak Belle. His recent works explore semantic foundations for neuro-symbolic weak supervision and the incorporation of logical constraints into deep generative models. More broadly, he is interested in developing AI systems that are more reliable, interpretable, and capable of reasoning under uncertainty. — 🗨️ ​Join us for an interactive session exploring Neuro-Symbolic AI, the emerging paradigm that blends the strengths of neural networks with symbolic reasoning. We will discuss how hybrid approaches can enhance generalization, interpretability, and reasoning, and how these methods are shaping the future of intelligent systems. Whether you’re a researcher, engineer, or simply curious about the cutting edge of AI, you’ll find an engaging space to learn, connect, and exchange ideas. Event Format: 🔍 Reading Group 🎙️ Panel Discussion 🧩 Tutorials/Workshop Series 💬 Research Roundtable 🛠️ Open-Source Project Review 📄 Paper Pitch 🤝 Collaboration Hour 🍵 Networking Mixer #AI #NeuroSymbolic #FutureOfAI