My interests include AI safety and its applications to the sciences. Some of my past research has been on MCMC, stochastic modeling, and computational biology
Life, 2020
We lay an analytical foundation for the large scale comparison of minerals and metalloenzyme ligands using chemoinformatics and molecular substructure search.
Natural Hazards, 2020
We study the stochastic characteristics of ocean environments and introduce stopping time into storm surge analysis.
Universe, 2021
We study a correlation between the planet mass and stellar metallicity of low-mass exoplanets orbiting spectral class G, K, and M stars.
The Planetary Science Journal, 2022
We present an investigation of hydrocarbon and nitrile species in Titan's upper atmosphere using stellar occultation observations obtained by Cassini/UVIS.
China Ocean Engineering, 2021
We introduce an improved parameter estimation method for extreme ocean environment based on moment estimation and the maximum entropy principle.
IEEE Conference on Bioinformatics and Computational Biology, 2021
We build machine learning models to predict enhancer-promoter interactions based on genomic features and extract aberrant features.
American Geophysical Union, 2020
American Geophysical Union, 2021
We segment key surface landforms from large Mars images using fast CNNs on frequency domain representations.
American Geophysical Union, 2020
The geological world is closely associated with biological history and function. Here, we present a mathematical method for comparing structural and chemical similarities between mineral-ligand pairs using molecular similarity metrics.
ICML 2024 Workshop on Structured Probabilistic Inference & Generative Modeling, 2024
We introduce a policy-gradient algorithm for optimizing the temperature ladder in parallel tempering MCMC. Our method outperforms SoTA for benchmark distributions.
arXiv:2606.04486, 2026
We propose a watermark for masked diffusion language models based on a vector-valued sketch of the text. Our method avoids dependency on local context, which is a common hook for erasure and imitation attacks.
A collection of my favorite puzzles.
Seven prisoners are each assigned a hat of seven possible colors, with color repetitions allowed. Every prisoner can see the hat colors of the other six prisoners, but not their own. They cannot communicate with others in any form, or else they are immediately executed. Then each prisoner writes down his guess of his own hat color. If at least one prisoner correctly guesses the color of his hat, they all will be set free immediately; otherwise they will be executed. They are given the night to come up with a strategy. Is there a strategy that they can guarantee that they will be set free?
One hundred ants are placed on a meter stick and simultaneously begin walking left or right at 1 cm/second. When two ants collide, they both reverse direction. If an ant reaches the end of the stick, it falls off. What arrangement of ants maximizes the time before all ants have fallen off? How long can they last?
Consider an equilateral triangle inscribed in a circle. Suppose a chord of the circle is chosen at random. What is the probability that the chord is longer than a side of the triangle?
A jailer places a coin on each square of a chessboard, heads or tails as he pleases, then hides a key under one square, which he reveals to the first prisoner. The first prisoner must flip exactly one coin and leave. The second prisoner then enters, sees only the arrangement of coins, and must name the square hiding the key. The prisoners may agree on a strategy beforehand. Can they guarantee escape?
Checkers occupy every square below an infinite horizontal line. As in peg solitaire, a checker may jump horizontally or vertically over an adjacent checker into an empty square, removing the checker that was jumped. How far above the line can a checker advance?
Countably many prisoners are each assigned a red or blue hat. Every prisoner can see the hats of all the others, but not their own, and no communication of any kind is allowed. All prisoners simultaneously guess their own hat color. They may agree on a strategy the night before. Is there a strategy under which only finitely many prisoners guess incorrectly?
Ten points are drawn on a table, arranged however an adversary likes. Show that they can always be covered using identical circular coins that do not overlap. (The coins may hang off the table, and fewer than ten may be used.)