Zishen Wan (Georgia Tech) has made significant contributions to the field of autonomous machines and cognitive systems, through vertically integrated application discovery, systems thinking, and software-hardware co-design intelligence. Several of his projects received recognition for their innovation and impact, including the recent award-winning CogSys, H3DFact, MulBERRY, and AutoPilot, which pushed the boundaries of neurosymbolic AI and autonomous intelligence, achieving impressive real-time efficiency and scalability. Notably, the development of CogSys, a reconfigurable neurosymbolic architecture that enables 75x speedup over traditional ML architectures with minimal area overhead, marks a milestone in neurosymbolic hardware. Zishen’s achievements have earned him accolades such as the 2023 Machine Learning Systems Rising Star, the 2024 Cyber-Physical Systems Rising Star, and multiple Best Paper Awards.
1. Tell us a bit about your background. What led you to pursue your current field of study?
My research lies at the intersection of computer architecture, VLSI, and embedded system, focusing on software-system-hardware co-design for autonomous machines and embodied intelligence, with the vision to advance their performance, efficiency, resilience, and trustworthiness.
While we enjoy the benefits of AI advancements, its development is on an unsustainable trajectory and lack of robustness and interpretability. We have yet to see AI agents that reason and make decisions in a reliable and explainable manner, robots that seamlessly interact with humans in complex environments, and intelligence that is pervasive yet untethered from the cloud. These challenges motivate my exploration of designing autonomous and embodied intelligence computing systems to address these gaps and push the boundaries of what AI can achieve.
2. How does your research focus support the mission of CoCoSys?
My research adopts a vertically integrated approach to co-design cognitive autonomous and embodied systems, seamlessly integrating CoCoSys tasks to improve real-time responsiveness, energy efficiency, explainability, and trustworthiness.
In our recent efforts, we systematically review and categorize neuro-symbolic-probabilistic algorithms (ISPASS’24) [Theme 1], analyze their computing system characteristics (TCASAI’24) and propose reconfigurable neurosymbolic architecture and dataflow (HPCA’25) [Theme 2], leverage CMOS and emerging memory technologies to design hardware motifs (DATE’24) [Theme 3], and explore efficiency-performance-robustness co-optimization (ASPLOS’24, Communication of ACM’24) and benchmark autonomous intelligence (ICRA’24) [Theme 4].Through these contributions, my research resonates with CoCoSys’s mission of advancing cognitive system design and innovation.
3. What’s a recent accomplishment or project you are particularly proud of?
Our software-system-hardware co-design for autonomous and cognitive intelligence research has been recently recognized by the community: H3DFact won Best Presentation Award at 2024 SRC TECHCON, CogSys won Best Poster Award at 2024 CoCoSys annual review, MulBERRY won Best Poster Award at 2023 IBM IEEE AI Compute Symposium, RobotPerf won Best Paper Award at 2023 IROS workshop. I am also honored as 2023 Machine Learning Systems Rising Star and 2024 Cyber-Physical Systems Rising Star.
Particularly, we recently develop an algorithm-hardware co-design framework, CogSys, which to the best of our knowledge is the first to achieve real-time efficiency and scalability of cognitive neurosymbolic systems. It features neuro/symbolic processing elements, dataflows, mapping strategies, and adaptive scheduler, enabling reconfigurable support for neural and symbolic kernels. It demonstrates over 75x speedup compared to TPU-like systolic array, with less than 5% area overhead, making it highly deployable and paving the way for neurosymbolic AI development. This work will appear in HPCA’25.
4. What have you learned from your experience as a CoCoSys research scholar advised by director Arijit Raychowdhury and Theme 2 Co-leader Tushar Krishna?
Working closely with my advisors Prof. Raychowdhury and Krishna, as well as collaborating with CoCoSys professors Lin, Naeemi, Panda, Rabaey, Raghunathan, Roy, Tenenbaum and their students has been an incredible privilege. Their mentorship, along with insights from fellow CoCoSys scholars, has significantly shaped my research and growth as a scholar.
Key lessons I’ve learned include: (1) Visionary Thinking: Tackle important and impactful problems, aim high, and embrace honest feedback. (2) Cross-Disciplinary Integration: Innovate through application discovery, systems thinking, and co-design across algorithm, system, and hardware layers. (3) Attention to Detail: Clearly articulate complex ideas through effective writing and presentations. (4) Kindness and Adaptability: Foster supportive teamwork, emphasizing collaboration and shared success.
Through their mentorship, I’ve gained technical expertise and a deeper understanding of what it means to be a thoughtful researcher and leader. Their guidance will continue to shape my career and aspirations.
5. What are your short-term and long-term career goals?
In the short term, I aim to build energy-efficient, robust, and trustworthy autonomous and embodied agent systems by synergizing efforts in neural, symbolic, and probabilistic algorithms, algorithm-hardware co-design, technology-driven hardware motifs, and collective intelligence.
In the long term, I aspire to advance the field of cognitive system design in academia, driving innovation and mentoring future researchers to push the boundaries of autonomous machines and cognitive AI computing.
6. For fun: If you could have any technology-related superpower, what would it be and why?
If I could have any technology-related superpower, it would be the ability to instantly understand and solve any technological problem to create positive change in society. With this superpower, I could tackle pressing global challenges like designing sustainable energy systems, improving healthcare technologies, or enhancing accessibility for people with disabilities.
For example, I could revolutionize clean energy solutions to combat climate change, optimize healthcare technologies for early diagnosis and treatment, or develop AI systems that provide equitable access to education worldwide. This power would let me accelerate technological innovation where it’s needed most, ensuring that advancements directly benefit humanity and help create a more equitable, sustainable, and inclusive future.