Synthesis-Enabled Reasoning: Towards a New Era of Formal Verification
11:40 AM to 1:00 PM
CSB 451 CS Auditorium
Adwait Godbole, UC Berkeley
Abstract:
Bugs in hardware and software are ubiquitous and costly, and AI-based development threatens to make them worse. While formal reasoning can prevent these failures by providing rigorous correctness guarantees, a critical bottleneck limits its adoption: the effort required to develop verification-friendly models. This effort involves synthesis tasks: decomposing the system, developing cross-layer specifications, and reformulating properties to be verification-amenable. Traditionally, these tasks have been performed manually and necessitate dual expertise in both systems and formal methods.
My research develops synthesis-enabled reasoning, a paradigm that automates these synthesis tasks, and applies it to hardware-software verification. Specifically, I demonstrate improved (1) scalability by automatically synthesizing compositional abstractions from RTL microarchitecture; (2) completeness by automatically generating attack-detection patterns, which discovered patterns missed by manual approaches; and (3) applicability by automatically converting axiomatic specifications to operational models, enabling previously intractable verification. Synthesis-enabled reasoning leverages rapid improvements in search (e.g., using neural techniques), and is thus poised to transform formal reasoning from an expert-only specialty into routine engineering practice.
Add to GMail
Info Session: Chainalysis
1:00 PM to 2:00 PM
CESPR 750
Chainalysis is the blockchain data platform that provides data, software, services, and research to government agencies, exchanges, financial institutions, and insurance and cybersecurity companies in over 70 countries. Our data powers investigation, compliance, and market intelligence software that has been used to solve some of the world's most high-profile criminal cases and grow consumer access to cryptocurrency safely. They are actively recruiting for SWE new grads and internship roles. Representatives will conduct an Employer informational session and offer a presentation about their company, mission, past/upcoming projects, and future recruitment efforts, followed by a Q&A session. For more information regarding the company, please feel free to the company website: https://www.chainalysis.com/.
Registration Information will be posted via email and CampusGroups.
*Event Audience: CS Graduate Students (MS/PhD) & Bridge Students
Learning to design new molecular interactions
11:40 AM to 1:00 PM
CSB 451 CS Auditorium
Chloe Hsu, UC Berkeley
Abstract:
Generalization is a core challenge in AI for scientific discovery. How do we learn from existing data to discover something fundamentally new? Protein design is one such example. Despite recent breakthroughs in protein structure prediction, designing new protein interactions remains challenging. We developed “inverse folding” models to generate protein sequences based on 3D structures. Surprisingly, we demonstrate that these models can successfully design dynamic interactions while only being trained on static structures. This seminar will discuss the mechanism for such generalization as inspiration for new machine learning algorithms in biology and beyond.
Reliable and Actionable AI for Healthcare
11:40 AM to 1:00 PM
CSB 451 CS Auditorium
Sheng Liu, Stanford University
Abstract:
While modern AI models achieve impressive performance on standard benchmarks, they remain brittle in high-stakes domains such as healthcare, where uncertainty is pervasive and errors are costly. The central challenge is not model capacity, but bridging the gap between predictive accuracy in controlled settings and reliable behavior in real-world clinical workflows. Existing methods often struggle with noisy, multimodal real-world data and lack principled mechanisms for incorporating expert knowledge during deployment.
In this talk, I present a set of machine learning methods designed to address these challenges. I will first describe approaches that leverage training dynamics and implicit signals to learn robust models from uncurated, imperfect data without amplifying noise. I will then introduce an inference-time framework that treats natural language not merely as input, but as a control interface, allowing domain experts to steer model behavior and agentic systems through linguistic feedback. By integrating robust learning and language-based steering within agentic AI systems, I demonstrate how AI can operate in closed-loop clinical settings with transparent reasoning, adaptive behavior, and human oversight.
Virtual Info Session: Prospect Equities
2:00 PM to 3:00 PM
Zoom
Prospect Equities is a full service real estate company offering brokerage and consulting services in residential and commercial real estate, apartment rentals, business brokerage, and builder/developer representation with hundreds of real estate professionals licensed throughout our offices in IL, NY & FL. They are actively recruiting for various internship roles. Representatives will conduct a Virtual Employer informational session and offer a presentation about their company, mission, past/upcoming projects and future recruitment efforts followed by a Q&A session.For more information regarding the company, please feel free to the company website: https://www.prospectequities.com/.
Registration Information will be posted via email, VMock and CampusGroups.
*Event Audience: CS Graduate Students (MS/PhD) & Bridge Students.
Add to GMail