Efficient Processing of Neuro-Symbolic AI: A Tutorial and Cross-Layer Co-Design Case Study

Zishen Wan, Che-Kai Liu, Hanchen Yang, Ritik Raj, Arijit Raychowdhury, Tushar Krishna
Proceedings of the International Conference on Neuro-symbolic Systems, PMLR 288:490-504, 2025.

Abstract

While neural-based models have driven recent breakthroughs in artificial intelligence (AI), they face critical challenges in unsustainable computational demands, limited robustness, and lack of interpretability. Neuro-symbolic (NeSy) AI has emerged as a promising paradigm that integrates neural learning and symbolic reasoning to enhance explainability, robustness, and data efficiency. Recent NeSy systems demonstrated strong potential in reasoning and trustworthy decision-making tasks, making them particularly suitable for cognitive human-AI applications. This tutorial presents a vertically integrated approach on the efficient processing of NeSy AI, bridging workload characteristics with system and hardware co-design. We begin by systematically categorizing NeSy workloads and analyzing their computational and memory demands to expose performance bottlenecks and optimization opportunities. Building on these insights, we focus on a class of vector-symbolic architecture-based NeSy systems and present a series of hardware case studies, including processing element microarchitecture, dataflow, FPGA design, and system-on-chip prototype. Our results highlight the efficiency and scalability improvements of NeSy systems, with the integration of application discovery, systems thinking, and co-design intelligence. Project Website: https://effi-nesy.github.io.

Cite this Paper

BibTeX
@InProceedings{pmlr-v288-wan25a, title = {Efficient Processing of Neuro-Symbolic AI: A Tutorial and Cross-Layer Co-Design Case Study}, author = {Wan, Zishen and Liu, Che-Kai and Yang, Hanchen and Raj, Ritik and Raychowdhury, Arijit and Krishna, Tushar}, booktitle = {Proceedings of the International Conference on Neuro-symbolic Systems}, pages = {490--504}, year = {2025}, editor = {Pappas, George and Ravikumar, Pradeep and Seshia, Sanjit A.}, volume = {288}, series = {Proceedings of Machine Learning Research}, month = {28--30 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v288/main/assets/wan25a/wan25a.pdf}, url = {https://proceedings.mlr.press/v288/wan25a.html}, abstract = {While neural-based models have driven recent breakthroughs in artificial intelligence (AI), they face critical challenges in unsustainable computational demands, limited robustness, and lack of interpretability. Neuro-symbolic (NeSy) AI has emerged as a promising paradigm that integrates neural learning and symbolic reasoning to enhance explainability, robustness, and data efficiency. Recent NeSy systems demonstrated strong potential in reasoning and trustworthy decision-making tasks, making them particularly suitable for cognitive human-AI applications. This tutorial presents a vertically integrated approach on the efficient processing of NeSy AI, bridging workload characteristics with system and hardware co-design. We begin by systematically categorizing NeSy workloads and analyzing their computational and memory demands to expose performance bottlenecks and optimization opportunities. Building on these insights, we focus on a class of vector-symbolic architecture-based NeSy systems and present a series of hardware case studies, including processing element microarchitecture, dataflow, FPGA design, and system-on-chip prototype. Our results highlight the efficiency and scalability improvements of NeSy systems, with the integration of application discovery, systems thinking, and co-design intelligence. Project Website: https://effi-nesy.github.io.} }
Endnote
%0 Conference Paper %T Efficient Processing of Neuro-Symbolic AI: A Tutorial and Cross-Layer Co-Design Case Study %A Zishen Wan %A Che-Kai Liu %A Hanchen Yang %A Ritik Raj %A Arijit Raychowdhury %A Tushar Krishna %B Proceedings of the International Conference on Neuro-symbolic Systems %C Proceedings of Machine Learning Research %D 2025 %E George Pappas %E Pradeep Ravikumar %E Sanjit A. Seshia %F pmlr-v288-wan25a %I PMLR %P 490--504 %U https://proceedings.mlr.press/v288/wan25a.html %V 288 %X While neural-based models have driven recent breakthroughs in artificial intelligence (AI), they face critical challenges in unsustainable computational demands, limited robustness, and lack of interpretability. Neuro-symbolic (NeSy) AI has emerged as a promising paradigm that integrates neural learning and symbolic reasoning to enhance explainability, robustness, and data efficiency. Recent NeSy systems demonstrated strong potential in reasoning and trustworthy decision-making tasks, making them particularly suitable for cognitive human-AI applications. This tutorial presents a vertically integrated approach on the efficient processing of NeSy AI, bridging workload characteristics with system and hardware co-design. We begin by systematically categorizing NeSy workloads and analyzing their computational and memory demands to expose performance bottlenecks and optimization opportunities. Building on these insights, we focus on a class of vector-symbolic architecture-based NeSy systems and present a series of hardware case studies, including processing element microarchitecture, dataflow, FPGA design, and system-on-chip prototype. Our results highlight the efficiency and scalability improvements of NeSy systems, with the integration of application discovery, systems thinking, and co-design intelligence. Project Website: https://effi-nesy.github.io.
APA
Wan, Z., Liu, C., Yang, H., Raj, R., Raychowdhury, A. & Krishna, T.. (2025). Efficient Processing of Neuro-Symbolic AI: A Tutorial and Cross-Layer Co-Design Case Study. Proceedings of the International Conference on Neuro-symbolic Systems, in Proceedings of Machine Learning Research 288:490-504 Available from https://proceedings.mlr.press/v288/wan25a.html.

Related Material