9th Annual Simons Institute Industry Day
2023-11-02
Berkeley, United States
as the organisers put it
The Simons Institute’s ninth Industry Day was our largest to date, with over 150 attendees from the Institute, the broader UC Berkeley campus, our partner and sponsor companies, and beyond. The event, which took place on November 2, was designed to facilitate knowledge-sharing among industry and academic scientists, and to highlight the importance of industry partners in supporting research in the foundations of computing.
who it's for
research scientists in industry and academia, algorithms engineers, database engineers and architects, machine learning researchers, cryptographers and security engineers, quantum computing researchers, data scientists
topics
large language models and transformers; hypergraph databases; incremental view maintenance; sketching and sampling for queries; quantum LDPC codes; fault-tolerant quantum algorithms; sublinear algorithms and property testing; privacy-preserving join algorithms; probabilistic circuits for trustworthy AI; non-monotonic gradient descent dynamics; strongly polynomial 2-sparse linear programming; automated program verification with large models
| title | author | length |
|---|---|---|
| Introduction and Program Highlights | Sandy Irani | 04m |
Ninth Annual Industry Day presents an overview of current and planned semester-long programs, naming a new focus on large language models and transformers that… | ||
| Logic and Algorithms In Database Theory and AI | Hung Ngo | 04m |
Frames a database as a collection of hypergraph databases and maps concrete problems — triangles as conjunctive queries, APSP as a recursive logical query — to… | ||
| Error-Correcting Codes: Theory and Practice | Venkat Guruswami | 04m |
A semester-long program on error-correcting codes brings algebraic codes achieving capacity (worst-case and random models), rapid advances in polar codes in 5G… | ||
| Quantum Algorithms, Complexity, and Fault Tolerance | Sandy Irani | 03m |
Theory reports that hardware now reaches 1,000-plus qubits but qubit error rates force focus on fault tolerance as the next milestone. Recent constructions… | ||
| Sublinear Algorithms | Jelani Nelson | 03m |
Program runs May 20–August 9 2024 and lists 29 long-term participants with applications still open. Core themes are streaming and sketching algorithms… | ||
| Modern Paradigms in Generalization | Matus Telgarsky | 10m |
Program sets out to produce a taxonomy of settings and to adapt old proofs while building new tools for non IID generalization, with concrete activities: four… | ||
| Cryptography 10 Years Later: Obfuscation, Proof Systems, and Secure Computation | Abhishek Jain | 10m |
A summer 2025 Simons program focuses on three concrete advances and remaining gaps: general-purpose obfuscation from well-studied assumptions exists but… | ||
| Vertex Connectivity in Poly-logarithmic Max-flows | Sorrachai Yingchareonthawornchai | 07m |
Proves vertex connectivity (global and s–t) using only poly-logarithmic max-flows instead of the previous ~n max-flow runs, by generating many sublinear-size… | ||
| Join Algorithms Meet New Constraints: Scalability, Latency and Privacy | Xiao Hu | 07m |
Algorithms that scale joins across hundreds or thousands of machines by minimizing communication rounds and per-server load, with formal claims of an optimal… | ||
| Probabilistic Reasoning and Learning for Trustworthy AI | YooJung Choi | 07m |
Frames fairness, robustness and explainability as probabilistic inference problems and builds a backend that actually answers those queries under uncertainty… | ||
| Non-monotonic convergence of gradient descent with large stepsize | Jingfeng Wu | 05m |
Shows that for a simplified but rigorous setting — linear models on perfectly separable data — gradient descent with any constant step size provably drives… | ||
| How Close is Close Enough? Solving Linear Programs When Rounding Just Won't Cut It! | Bento Natura | 05m |
Shows that linear programs whose constraint normals are 2-sparse constraints admit a new algorithm that runs in strongly polynomial many arithmetic operations… | ||
| How looking at a problem quantumly could solve it faster? | Hamoon Mousavi | 06m |
Quantum version of constraint satisfaction sometimes admits solutions where classical version provably doesn't: the 3x3 magic square has no ±1 assignment but… | ||
| Decoding the Maze: New Frontiers in Achieving Nash Equilibrium in AI Architectures | Manolis Vlatakis Gkaragkounis | 06m |
Claim: composing convex losses with neural nets creates a hidden game where parameter-to-action maps destroy convexity and introduce indirect competition… | ||
| From Robustness to Privacy and Back | Lydia Zakynthinou | 09m |
Gives a black-box transformation that turns any robust estimator into an ε‑differentially private estimator, with the DP error matching the robust error for a… | ||
| Well-Being, AI, and You | Alon Halevy | 44m |
Argues for an operating system for well-being that lets apps coordinate instead of competing, built from three components: database of you (a personal timeline… | ||
| Theory matters: how to build a successful company by trusting theoreticians | Molham Aref | 17m |
Argues that trusting database theory repeatedly pays off: a LogicBlox super-optimized datalog engine got built and remains heavily cited, relational systems… | ||
| DBSP: A unified (practical) theory of databases, streams, and incremental computation | Mihai Budiu, Leonid Ryzhyk | 18m |
DBSP presents a concrete claim: any SQL or Datalog query can be mechanically transformed into a streaming, incremental program by treating tables as streams… | ||
| Challenges in Research on Practical Algorithms | Andrew Goldberg | 26m |
Argues that practical algorithm research is shrinking in industry and gives concrete failures and fixes: adding a 100-second U-turn penalty to public road data… | ||
| The Data Science behind Cruise’s Robotaxi Fleet | Or Cohen | 19m |
Describes concrete systems Cruise uses: dense safety metrics derived from near collisions and human accident benchmarks, simulation test case sampling for… | ||
| Relational Programming | Remy Wang | 05m |
One-line relational joins and semiring rewrites turn into concrete systems: a datalog engine (birelational AI) automatically parallelizes and distributes… | ||
| Unique Challenges and Opportunities in Working with Residential Real Estate Data | Takamitsu Tanaka | 19m |
A midsize residential lending shop describes a concrete pipeline: extract text with OCR, locate relevant blocks with TFIDF and regression, then populate a… | ||
| Lemur: Integrating Large Language Models in Automated Program Verification | Nina Narodytska | 16m |
A framework named Lemur integrates GPT-4 via OpenAI API with model-checkers UAutomizer and ESBMC, using LLM calls to propose, strengthen and repair loop… | ||
| Introducing RelationalAI | Molham Aref | 14m |
RelationalAI positions itself as an AI coprocessor that embeds inside Snowflake as materialized views, avoiding data copies and preserving governance, and… | ||
| Market Algorithms for Autobidding | Gagan Aggarwal | 19m |
Autobidding is framed as an LP: items with known prices, a budget and TCPA constraint, and a practical algorithm solve dual on a sample that yields two dual… | ||
| Data Structures and Optimization for Fast Algorithms | Aaron Sidford | 03m |
Program surveys concrete advances that shave decades off runtimes for tasks like linear system solving, convex programming, flow and Markov decision process… | ||
| nothing matches | ||