Continuous LWE is as Hard as LWE
summary
Claim: continuous LWE (CLWE) is as hard as standard LWE via a classical reduction. Proof path: use LWE leakage-resilience to convert to binary sparse secrets (norm ≈ √K), smooth discrete A to continuous A with a Gaussian preimage sampling lemma (controls noise blowup ∝ secret norm), then apply random rotations to make the secret uniform on the unit sphere. Parameter map: γ ≈ secret norm, β ≈ σ/q. Consequences: classical worst-case lattice hardness transfers to CLWE and yields new hardness for learning periodic neurons, undetectable Fourier backdoors, and mixtures of Gaussians with ≈γ components (improved from √n to n^ε or log^{1+ε} under stronger LWE assumptions).
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:27Outline
- 0:50Learning With Errors Regev '05
- 1:56Why care about LWE?
- 7:41Continuous LWE [Bruna-Regev-Song-Tang '21] LWE
- 12:44What do we know about CLWE?
- 25:16Homogeneous CLWE (hCLWE)
- 26:06What else does CLWE get us? Theorem (BRST 21); CLWE implies hardness of learning mixtures of Oly
- 27:44LWE vs. CLWE
- 34:26Our Results Theorem Gupte-V.-Vaikuntanathan "22 : New hardness for density estimation for mixtures of Gaussians in R" in poly(n) time Number of Gaussian components in mixture
- 37:39Interpretation
- 38:26Our Results Theorem Gupte-V.-Vaikuntanathan "22 : New hardness for density estimation for mixtures of Gaussians in R" in poly(n) times Number of Gaussian components in mixture
- 38:59General Reduction From LWE to CLWE
- 43:51Example: Step 2
- 47:01Open Questions
related talks
47mHardness of LWE on General Entropic Distributions
30mOn the Concrete Security of LWE with Small Secret
57mLattice Assumptions with Hints: Succinct LWE and its Applications
01h 28mThe Learning With Errors Problem and Cryptographic Applications
01h 29mLattices, Multilinear Maps and Program Obfuscation
01h 07mLattices and Encrypted Computation
01h 04mThe Mathematics of Lattices II
37mAlgebraic Algorithms for LWE
