This post explains the key findings and algorithm of our 2023 SIGGRAPH paper “Winding Numbers on Discrete Surfaces”.
For a given point on a surface, we would like to know whether it is inside or outside a given curve.
For example, coloring points on this cow according to inside…
In this post, I’ll be discussing my part of my work on how to use compiler
analysis to aid in optimizing memory allocation.
I will begin with some background and motivation, and then introduce my
proposed workflow as well as some interesting
challenges that this work involves.
C…
Large language models (LLMs) represent a promising but controversial assistant in the process of preparing and reviewing scientific papers. They may serve as useful analytical tools to scrutinize manuscripts and identify possible weaknesses or inaccuracies that need addressing. H…
The workload of state-vector quantum circuit simulation is massively parallel. Most quantum gates multiply one small matrix to many vectors, and the workload is balanced.
Challenge: the state vector is too large to be stored on one GPU, and the communication cost is very high. …
SQL is the de facto query language used to interact with databases. Since SQL is declarative, users express the intended result of their query rather than the concrete procedural steps to produce the answer. Database management systems (DBMSs) find fast execution strategies for …
In this blog post, I will begin by introducing the concept of cut sparsifier for a given graph
G, which has been a powerful tool in the design of graph algorithms. Following that, I will present a tight lower bound for the directed cut sparsifier of a balanced directed graph,…
Applications access their data in the memory using virtual addresses, which tell the hardware where to fetch or store a particular piece of data.
Virtual addresses are mapped to physical addresses referencing the actual memory provided by the hardware (i.e. main memory) in the …
In June, our research group released MLCEngine,
a universal LLM deployment engine powered by machine learning compilation.
MLCEngine is a single engine to enable LLM deployment across both cloud and edge devices, with full support for OpenAI API.
Performance has been one of the k…
Automated reasoning (AR) is a branch of artificial intelligence that applies various reasoning techniques to solve problems from mathematics and logic.
AR engines use clever optimizations and heuristics to solve typically intractable (e.g., NP-hard) problems.
Before a problem …
This post describes a part of our work NodeMedic-FINE, an automated tool to find arbitrary command injection and arbitrary code execution vulnerabilities in Node.js packages. While I am the sole author of this blog post, the work it describes was done in collaboration with Darion…
