Welcome to my page. I will get this place in order one day.

My name is Anton and I’m an undergrad graduate in math and compsci from Arizona State university. I focus broadly in probability theory and algorithms, but I like to learn about all sorts of things.

I occasionally play CTF(Capture the Flag) competitions with our local uni team, CTF Academy, as well as with Shellphish.

I speak Russian and English and at some point have also studied Spanish and Mandarin (to varying degrees of success).

Here’s some stuff I like, in no particular order

✶ Stochastic processes & models, statistical algos, information theory;

✶ Random, approximation, and reconstruction algorithms;

✶ Competitive programming (as of recently);

✶ Graph theory;

✶ Diffusion models, reinforcement learning;

✶ Cryptography & cryptanalysis;

✶ Computational biology;

What I’ve worked on recently

✫ Minimum entropy coupling steganography: this is an algorithm that allows one to embed arbitrary data within the sample path of a random signal, such that the distribution of the signal remains unaffected. For example, one can encode a message within the choice of tokens of an LLM - we manipulate the token selection, but can guarantee that the quality of the output is not worsened by this.

→ I built a web demo that you can try out here!, as well as an introductory write-up you can read.

✫ Trace reconstruction algorithms: This is a project for my randomized algorithms class. The problem is to reconstruct a string $x=x_1,...,x_n$ given a set of traces, where a trace is a subsequence of $x$ in which each character is deleted randomly with fixed probability $p$. The main questions are: 1) how many traces does it take to rebuild $x$ with good accuracy, and 2) how to do it at all.

→ Read the complete write up here.

✫ DeepMapDB: this has been a long-running on-and-off r&d project on the following question: How much data can you fit in a neural network? Put more specifically, given a dictionary-format database, what is the smallest NN that can “memorize it” under fixed train time constraints and up to near-perfect accuracy? This task has applications in scenarios in plain data compression, DMBS with high-volume querying requirements, and database deployment on space and memory-constrained hardware.

✫ crypto.college: my university uses pwn.college, a platform for hosting CTF-style challenges to teach its cybersec classes. Check it out, it’s brilliant. Though security such as low-level exploitation is not really my forte, I have some knowledge of cryptography from playing CTFs and personal research. I’m currently working with a team on expanding pwn.college to include several modules of cryptography material & challenges, which will be hopefully used one day to teach classes at ASU!