On the Assumptions Used for Obfuscation
summary
Survey compares four concrete assumptions used to build obfuscation: LWE, SXDH, local PRGs, and LPN. Gives concrete regimes for LPN (constant noise + linear samples → exponential best attacks; inverse‑poly noise → subexponential attacks via low‑weight kernel vectors and BKW variants), proves decision→search via Goldreich‑Levin, and describes Zq/LPN lifting that lets one “plant” a short kernel vector for CCA‑style keys. Shows local PRGs from sparse LPN require expansion and resilient, high‑degree predicates, and orders primitives by available homomorphism: none → bounded addition → bilinear → full.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:48Obfuscation
- 4:24Learning Parity with Noise [BFKL94]
- 6:15Decoding Random Linear Code GKLE
- 13:52Known Attacks
- 19:09Simple Distinguishing Attack
- 24:33Pseudorandomness
- 30:51Random Self-Reducibility
- 31:22Dual Version: Syndrome Decoding
- 32:44Corollary: Planting Short Vector in Kerne
- 34:02LPN: Evidence for Hardness
- 38:17LPN: Features
- 40:47Variants
- 52:59Learning with Errors Variant Regevos
- 54:05Locally Computable Functions (NC) Each output depends on constant number of inputs
- 54:46Lin-PRGs in NC
- 56:53Poly-Stretch PRG in NC
- 59:09Generalization to Long Output
related talks
46mIndistinguishability Obfuscation from Bilinear Maps and LPN Variants
01h 04mTutorial on obfuscation
01h 26mObfuscation 1
58mObfuscation 2
01h 08mExpedition to Obfustopia: Indistinguishability Obfuscation from Well-Studied Assumptions to New...
38mRecent Advances on Foundations of Program Obfuscation
01h 29mLattices, Multilinear Maps and Program Obfuscation
01h 05mObfuscation from Noisy Linear FE
