summary | arXiv | program | project page | code | video

We present Accel, a novel corrective fusion network that combines (1) optical flow-based feature warping with (2) lightweight, per-frame, temporal correction to achieve state-of-the-art accuracy and throughput on video semantic segmentation.

summary | arXiv | program | website

Enterprises are increasingly deploying large camera networks for video analytics. But video analytics is compute and data intensive, with cost growing with the number of cameras and operation time. To address this cost challenge, we present Spatula, a new system for efficient cross-camera video analytics. Spatula exploits spatial and temporal locality in the dynamics of real camera networks to guide its inference-time search for a query identity.

summary | arXiv | program | website | video

We discuss the potential of spatio-temporal correlations -- content correlations between geographically proximate cameras in wide-area enterprise camera deployments -- to improve cost efficiency and inference accuracy in large-scale video analytics operations. Our template application is real-time person re-identification and tracking.

summary | arXiv | program

We exploit video compression techniques (in particular, the block motion vectors in H.264 video) and feature similarity across frames to accelerate a classical scene recognition task, semantic segmentation, on video.

summary | pdf | program

We investigate the fundamental tradeoff between exposure to online (network-based) and offline threats faced by a Bitcoin exchange that must store Bitcoin across online and offline storage, while guaranteeing availability to customers. Parameterizing deposit, withdrawal, and theft events as Poisson processes, we are able to model the financial dynamics of the exchange, and solve for the optimal threshold on online storage.

summary | pdf

This body of work is motivated by a simple statement: machine learning systems must meet the performance requirements of the applications they enable. Advances in deep learning applied to vision have unlocked opportunity in robotic navigation, industrial and agricultural monitoring, and retail intelligence, each use case with its own latency, throughput, and cost constraints. This thesis is a step toward solving this constrained optimization problem.

summary | pdf | slides

Portal is a payment protocol and software system that enables one-click purchases of long-form news content on the Internet, without requiring a user to sign up for a subscription or login to a content provider's website.

summary | pdf | github

We propose a Bitcoin micropayments-based revenue system for online businesses, which enables users to make small payments to access web content on a per-use basis, in lieu of viewing ads or signing up for a credit card subscription.

summary | pdf | github

We investigate the fundamental tradeoff between exposure to online (network-based) and offline threats faced by a Bitcoin exchange that must store Bitcoin across online and offline storage, while guaranteeing availability to customers. Parameterizing deposit, withdrawal, and theft events as Poisson processes, we are able to model the financial dynamics of the exchange, and solve for the optimal threshold on online storage.