Source author record

Luke Anderson

Luke Anderson appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Cryptography and Security Distributed, Parallel, and Cluster Computing Graphics Machine Learning physics.comp-ph

Catalog footprint

What is connected

2works

5topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2020arXiv

DiffTaichi: Differentiable Programming for Physical Simulation

We present DiffTaichi, a new differentiable programming language tailored for building high-performance differentiable physical simulators. Based on an imperative programming language, DiffTaichi generates gradients of simulation steps using source code transformations that preserve arithmetic intensity and parallelism. A light-weight tape is used to record the whole simulation program structure and replay the gradient kernels in a reversed order, for end-to-end backpropagation. We demonstrate the performance and productivity of our language in gradient-based learning and optimization tasks on 10 different physical simulators. For example, a differentiable elastic object simulator written in our language is 4.2x shorter than the hand-engineered CUDA version yet runs as fast, and is 188x faster than the TensorFlow implementation. Using our differentiable programs, neural network controllers are typically optimized within only tens of iterations.

preprint2016arXiv

New kids on the block: an analysis of modern blockchains

Half a decade after Bitcoin became the first widely used cryptocurrency, blockchains are receiving considerable interest from industry and the research community. Modern blockchains feature services such as name registration and smart contracts. Some employ new forms of consensus, such as proof-of-stake instead of proof-of-work. However, these blockchains are so far relatively poorly investigated, despite the fact that they move considerable assets. In this paper, we explore three representative, modern blockchains---Ethereum, Namecoin, and Peercoin. Our focus is on the features that set them apart from the pure currency use case of Bitcoin. We investigate the blockchains' activity in terms of transactions and usage patterns, identifying some curiosities in the process. For Ethereum, we are mostly interested in the smart contract functionality it offers. We also carry out a brief analysis of issues that are introduced by negligent design of smart contracts. In the case of Namecoin, our focus is how the name registration is used and has developed over time. For Peercoin, we are interested in the use of proof-of-stake, as this consensus algorithm is poorly understood yet used to move considerable value. Finally, we relate the above to the fundamental characteristics of the underlying peer-to-peer networks. We present a crawler for Ethereum and give statistics on the network size. For Peercoin and Namecoin, we identify the relatively small size of the networks and the weak bootstrapping process.