HELLO! I'M YANSONG FENG ▼
My research applies algorithmic number theory and combinatorics to post-quantum cryptography and theoretical computer science. Now I am focusing on LLM post-training and exploring the intersection of AI and cryptography like AI for security, Trustworthy LLM, Privacy-preserving machine learning and AI-assisted formal verification.
EDUCATION & WORK EXPERIENCE
-
– Present. Ph.D. Student, Academy of Mathematics and Systems Science (AMSS), Chinese Academy of Sciences (CAS). Academic Supervisor: Yanbin Pan.
-
– Present. AI4Crypto Research Intern, MiniMax.
VISITING EXPERIENCE
-
– . The Hong Kong Polytechnic University, Hong Kong. Hosted by Man Ho Allen Au.
-
– . Crypto Group, Aarhus University, Aarhus, Denmark. Hosted by Diego F. Aranha.
-
(one week). Laboratory of Mathematics Nicolas Oresme, University of Caen Normandy, Caen, France. Hosted by Abderrahmane Nitaj.
-
– . National University of Singapore, Singapore. Hosted by Divesh Aggarwal.
SERVICE
- Program Committee
- CT-RSA
- Subreviewer
- STOC, SODA, ASIACRYPT, PKC
I expect to graduate in 2027 and am seeking full-time opportunities in industry or as a postdoctoral researcher. Please feel free to contact me!
★ INTERN EXPERIENCE
-
Post-training & data synthesis
Built cryptography data synthesis and training pipelines for SFT, mid-training, and long-horizon RL, with benchmarks for cryptanalysis and cryptographic software verification in Lean.
-
AI for Scientific Discovery I:
Designed a multi-agent framework using M3.1-Flash-Preview that disproved Tokareva’s conjecture through an infinite family of counterexamples; key theorems were verified in Lean 4. Paper MathWorld
-
AI for Scientific Discovery II:
Conducted automated cryptanalysis across modern cryptographic schemes, identifying several previously unknown structural flaws and turning multiple theoretical attacks into practical breaks. For example, results include COLM0 ciphertext forgery with two encryption queries and SBC-VOLE secret-key recovery from a single known signature. See our Blog and Github.
★ PUBLICATIONS [DBLP] [Google Scholar]
PEER-REVIEWED PUBLICATIONS
-
New Asymptotic Results on Predicting Polynomial Congruential Generators
-
Better Bounds for Finding Fixed-Degree Isogenies via Coppersmith's Method
-
On Factoring and Power Divisor Problems via Rank-3 Lattices and the Second Vector
-
Computing Asymptotic Bounds for Small Roots in Coppersmith's Method via Sumset Theory
-
Small Public Exponent Brings More: Improved Partial Key Exposure Attacks against RSA
-
Embedding Integer Lattices as Ideals into Polynomial Rings
-
Partial Prime Factor Exposure Attacks on Some RSA Variants
-
Generalized Implicit Factorization Problem
MANUSCRIPTS
-
Subset Balancing and Generalized Subset Sum via Lattices
-
Revisit Unravelled Linearization with Ehrhart (quasi-)Polynomial
AI made significant contributions to the manuscripts below.
-
SVP∞ Is NP-Hard for Power-of-two Cyclotomic Ideals
-
Finding Integer Roots of Polynomials in Nearly Optimal Time
-
A counterexample to Tokareva's conjecture on sums of two bent functions
-
Faster SVP in Polynomial Space
-
SVP Is NP-Hard for Some Rank-2 Cyclotomic Modules
-
CVP Is NP-Hard for Principal Cyclotomic Ideals
-
Discrete Gaussian Sampling Meets BDGL Decoding: Solving the Shortest Vector Problem in 20.5596n+o(n) Time
-
Solving the Shortest Vector Problem in 20.7314n+o(n) Time via Discrete Gaussian Sampling on Superlattices
-
Jacobi Signatures, Revisited