• Publications
  • Influence
Large Scale Distributed Deep Networks
TLDR
This paper considers the problem of training a deep network with billions of parameters using tens of thousands of CPU cores and develops two algorithms for large-scale distributed training, Downpour SGD and Sandblaster L-BFGS, which increase the scale and speed of deep network training.
On the (im)possibility of obfuscating programs
TLDR
It is proved that obfuscation is impossible, by constructing a family of efficient programs that are unobfuscatable, in the sense that given any efficient program, the “source code” of that program can be efficiently reconstructed.
An Investigation of Practical Approximate Nearest Neighbor Algorithms
TLDR
This paper asks the question: can earlier spatial data structure approaches to exact nearest neighbor, such as metric trees, be altered to provide approximate answers to proximity queries and if so, how and why and introduces a new kind of metric tree that allows overlap.
Co-Training and Expansion: Towards Bridging Theory and Practice
TLDR
A much weaker "expansion" assumption on the underlying data distribution is proposed, that is proved to be sufficient for iterative co-training to succeed given appropriately strong PAC-learning algorithms on each feature set, and that to some extent is necessary as well.
An empirical study of learning rates in deep neural networks for speech recognition
TLDR
A new variant `AdaDec' is proposed that decouples long-term learning-rate scheduling from per-parameter learning rate variation, and was found to result in higher frame accuracies than other methods.
On Simulation-Sound Trapdoor Commitments
TLDR
A new, simpler definition for an SSTC scheme is presented that admits more efficient constructions and can be used in a larger set of applications, and how to construct S STC schemes from any one-way functions and based on specific number-theoretic assumptions is shown.
Strengthening Zero-Knowledge Protocols Using Signatures
TLDR
A novel technique to convert a large class of existing honest-verifier zero-knowledge protocols into ones with these stronger properties in the common reference string model, using a signature scheme existentially unforgeable against adaptive chosen-message attacks.
Resource Fairness and Composability of Cryptographic Protocols
TLDR
This model specifies the ideally fair functionality as allowing parties to “invest resources” in return for outputs, but in such an event offering all other parties a fair deal, to avoid a well-known impossibility result for fair multi-party computation with corrupted majority.
On the (im)possibility of obfuscating programs : (Extended abstract)
Informally, an obfuscator O is an (efficient, probabilistic) compiler that takes as input a program (or circuit) P and produces a new program O(P) that has the same functionality as P yet is
Alternatives to Non-malleability: Definitions, Constructions, and Applications (Extended Abstract)
TLDR
It is shown that tag-based non-malleability (tnm) suffices to construct an efficient universally-composable secure message transmission (SMT) protocol, for which the only previ- ous solution was based on a public key encryption functionality whose security is equivalent to non- malleability.
...
...