Obfuscation II
summary
Candidate obfuscator compiles circuits into a matrix branching program randomized by Kilian randomization and protected by multilinear encodings and per-input alphas. Proof rules out algebraic attacks by mapping each algebraic monomial to a unique input, testing each via the Schwartz‑Zippel lemma, and relying on a new multilinear subgroup elimination assumption to hybrid-split inputs. Construction yields IO for NC1 (and via further assumptions to larger classes), but requires large fields, subexponential hardness and huge concrete blowups (astronomical additive factors like ~2^100).
AI-written summary · about
chapters (11)
- 0:00Obfuscation Construction
- 2:01Security vs. Algebraic Attacks
- 10:50Unbounded Simulation vs. Algebraic Attacks GGHRSW, BR, BGKPS, MSW
- 25:12Warm-up (Special-Case) Assumption k=1 Mmap over composite N, with 2 large prime factors: - One "special" prime factor
- 31:04The Assumption (GLW14): Multilinear Subgroup Elimination k Mmap over composite N, with many large prime factors
- 33:07How to argue security • We need proof of indistinguishability: iO(C) to iO(C)
- 34:55Overall reduction strategy
- 38:05Lesson: Ability to make this (minor) type of change is actually sufficient!
- 38:20Hybrids intuition
- 50:36Ways forward • Gain confidence in Multilinear Subgroup Elimination Assumption over Mmap candidates by attack/defend cycle. successful examples from past: block ciphers (AESI. Bilinear assumptions over specific elliptic curve families)
- 54:16What are we leaving on the table?
related talks
01h 06mObfuscation: Past, Present, and Possible Futures
01h 04mCryptanalysis of Candidate Program Obfuscators
01h 04mHow to Encrypt a Functionality
51mObfuscation via GGH15 Encodings, 10 Years Later
01h 04mTutorial on obfuscation
56mToward Obfuscation from Affine Determinant Programs
01h 26mObfuscation 1
01h 06mTowards general-purpose program obfuscation via local mixing
