sumcheck-builders-original.md

Sumcheck Builders Group Seminars

Meeting Link

Zoom Meeting Meeting ID: 913 291 7498 Passcode: 598160

Upcoming Talks

• Giacomo Fenzi from EPFL to talk about his work on STIR: Reed–Solomon Proximity Testing with Fewer Queries and WHIR: Reed–Solomon Proximity Testing with Super-Fast Verification

  • Two joint seminars with Nexus Speaker Series
  • STIR session:
    • Wendesnday, Oct 16, 11:00 AM PST / 2:00 PM EST / 8:00 PM CET
    • Meeting room: https://meet.google.com/zdp-vjud-aji
    • Recording
  • WHIR session:
    • Friday, Nov 8, 10:00 AM PST / 1:00 PM EST / 7:00 PM CET
    • Meeting room: The usual Zoom room

• Ron Rothblum from Succinct to talk about Blaze: Fast SNARKs from Interleaved RAA Codes.

  • Friday, Nov 1, 11:00 AM PST / 2:00 PM EST / 8:00 PM CET

Past Talks

• Simon Lau from Polyhedra: a presentation on his recent work in accelerating SumChecks using GPUs. Simon is a researcher specializing in efficient hardware and computer architectures. He will share his latest findings and benchmark results, comparing the performance of both CUDA and Triton frameworks in this context

  • Friday, October 18, 10:00 AM PST / 1:00 PM EST / 7:00 PM CET
  • Recording

• Vishruti (cryptography at %) will be talking about an optimization (dubbed "Beta Cascade") to the prover's sumcheck evaluation function over multilinear extensions of the form $V(g) = \sum_z \beta(g, z) * f(z)$ (where $f$ is nonlinear in each of the $z_i$'s), as is used in GKR over circuits with "structured" wiring patterns. In particular, Beta Cascade boasts an $O(2^n)$ per-sumcheck-round runtime ($n$ = number of "free" variables within the MLE), while requiring only $O(n)$ memory (vs. $O(2^n)$ for a "beta table" of evaluations) and an "update" step of $O(n)$ (vs. $O(2^n)$ for the "beta table" approach).

  • Thurdays, Oct 3, 1 PM PT/ 4 PM ET / 10 PM CET
  • Recording, Passcode: 7s?MHjHv

• Tiancheng from Polyhdera team is going to talk about Pianist.

  • Friday, Sept 27, 1 PM PT/ 4 PM ET / 10 PM CET
  • Abstract: Blockchain technology has grown rapidly, but scalability remains a major challenge. Zero-knowledge proofs (ZKPs), especially zkRollups, offer solutions but often require massive resources. In this talk, we introduce a fully distributed approach to ZKP generation using multiple machines with minimal communication. Built on the Plonk system, our protocols can be used as a recursive proof system, significantly reducing prover time and resource demands, offering a more scalable solution for blockchain applications.
  • Recording, Passcode: 9KV#RkJa

• Lev from the PSE team is sharing his method for turbocharging Keccak circuits with GF2_128 field. He's talking about hitting speeds of 60K Keccak per second - pretty wild!

  • Friday, Sept 13, 8 AM PT / 11 AM EST / 5 PM CET / 11 PM SGT
  • Recording, Passcode: #RjuB+Y5
  • Slides

• Zhenfei from the Polyhedra team is gonna talk about some cool tricks they've used in their Expander prover, including stuff from GKR$^2$ to accelerate Poseidon hash functions.

  • Friday, Sept 6, 8 AM PT / 11 AM EST / 5 PM CET / 11 PM SGT
  • Recording, Passcode: 4YYnVV1!