Matthai Philipose

ICML Conference 2021 Conference Paper

Boosting the Throughput and Accelerator Utilization of Specialized CNN Inference Beyond Increasing Batch Size

• Jack Kosaian
• Amar Phanishayee
• Matthai Philipose
• Debadeepta Dey
• Rashmi Vinayak

Datacenter vision systems widely use small, specialized convolutional neural networks (CNNs) trained on specific tasks for high-throughput inference. These settings employ accelerators with massive computational capacity, but which specialized CNNs underutilize due to having low arithmetic intensity. This results in suboptimal application-level throughput and poor returns on accelerator investment. Increasing batch size is the only known way to increase both application-level throughput and accelerator utilization for inference, but yields diminishing returns; specialized CNNs poorly utilize accelerators even with large batch size. We propose FoldedCNNs, a new approach to CNN design that increases inference throughput and utilization beyond large batch size. FoldedCNNs rethink the structure of inputs and layers of specialized CNNs to boost arithmetic intensity: in FoldedCNNs, f images with C channels each are concatenated into a single input with fC channels and jointly classified by a wider CNN. Increased arithmetic intensity in FoldedCNNs increases the throughput and GPU utilization of specialized CNN inference by up to 2. 5x and 2. 8x, with accuracy close to the original CNN in most cases.

NeurIPS Conference 2018 Conference Paper

Heterogeneous Bitwidth Binarization in Convolutional Neural Networks

• Joshua Fromm
• Shwetak Patel
• Matthai Philipose

Recent work has shown that fast, compact low-bitwidth neural networks can be surprisingly accurate. These networks use homogeneous binarization: all parameters in each layer or (more commonly) the whole model have the same low bitwidth (e. g. , 2 bits). However, modern hardware allows efficient designs where each arithmetic instruction can have a custom bitwidth, motivating heterogeneous binarization, where every parameter in the network may have a different bitwidth. In this paper, we show that it is feasible and useful to select bitwidths at the parameter granularity during training. For instance a heterogeneously quantized version of modern networks such as AlexNet and MobileNet, with the right mix of 1-, 2- and 3-bit parameters that average to just 1. 4 bits can equal the accuracy of homogeneous 2-bit versions of these networks. Further, we provide analyses to show that the heterogeneously binarized systems yield FPGA- and ASIC-based implementations that are correspondingly more efficient in both circuit area and energy efficiency than their homogeneous counterparts.

ICML Conference 2016 Conference Paper

Analysis of Deep Neural Networks with Extended Data Jacobian Matrix

• Shengjie Wang 0001
• Abdel-rahman Mohamed
• Rich Caruana
• Jeff A. Bilmes
• Matthai Philipose
• Matthew Richardson
• Krzysztof J. Geras
• Gregor Urban

Deep neural networks have achieved great successes on various machine learning tasks, however, there are many open fundamental questions to be answered. In this paper, we tackle the problem of quantifying the quality of learned wights of different networks with possibly different architectures, going beyond considering the final classification error as the only metric. We introduce \emphExtended Data Jacobian Matrix to help analyze properties of networks of various structures, finding that, the spectrum of the extended data jacobian matrix is a strong discriminating factor for networks of different structures and performance. Based on such observation, we propose a novel regularization method, which manages to improve the network performance comparably to dropout, which in turn verifies the observation.

AAAI Conference 2014 Conference Paper

Efficient Object Detection via Adaptive Online Selection of Sensor-Array Elements

• Matthai Philipose

TIST Journal 2011 Journal Article

Introduction to the special issue on intelligent systems for activity recognition

• Daqing Zhang
• Matthai Philipose
• Qiang Yang

IJCAI Conference 2007 Conference Paper

• Shiaokai Wang
• William Pentney
• Ana-Maria Popescu
• Tanzeem Choudhury
• Matthai Philipose

Given sensors to detect object use, commonsense priors of object usage in activities can reduce the need for labeled data in learning activity models. It is often useful, however, to understand how an object is being used, i. e. , the action performed on it. We show how to add personal sensor data (e. g. , accelerometers) to obtain this detail, with little labeling and feature selection overhead. By synchronizing the personal sensor data with object-use data, it is possible to use easily specified commonsense models to minimize labeling overhead. Further, combining a generative common sense model of activity with a discriminative model of actions can automate feature selection. On observed activity data, automatically trained action classifiers give 40/85% precision/recall on 10 actions. Adding actions to pure object-use improves precision/recall from 76/85% to 81/90% over 12 activities.

ICRA Conference 2004 Conference Paper

Mapping and Localization with RFID Technology

• Dirk Hähnel
• Wolfram Burgard
• Dieter Fox
• Kenneth P. Fishkin
• Matthai Philipose

We analyze whether radio frequency identification (RFID) technology can be used to improve the localization of mobile robots and persons in their environment. In particular we study the problem of localizing RFID tags with a mobile platform that is equipped with a pair of RFID antennas. We present a probabilistic measurement model for RFID readers that allow us to accurately localize RFID tags in the environment. We also demonstrate how such maps can be used to localize a robot and persons in their environment. Finally, we present experiments illustrating that the computational requirements for global robot localization can be reduced strongly by fusing RFID information with laser data.

TCS Journal 2000 Journal Article

DyC: an expressive annotation-directed dynamic compiler for C

• Brian Grant
• Markus Mock
• Matthai Philipose
• Craig Chambers
• Susan J. Eggers

We present the design of DyC, a dynamic-compilation system for C based on run-time specialization. Directed by a few declarative user annotations that specify the variables and code on which dynamic compilation should take place, a binding-time analysis computes the set of run-time constants at each program point in the annotated procedure's control-flow graph; the analysis supports program-point-specific polyvariant division and specialization. The results of the analysis guide the construction of a run-time specializer for each dynamically compiled region; the specializer supports various caching strategies for managing dynamically generated code and mixes of speculative and demand-driven specialization of dynamic branch successors. Most of the key cost/benefit trade-offs in the binding-time analysis and the run-time specializer are open to user control through declarative policy annotations. DyC has been implemented in the context of an optimizing compiler, and initial results have been promising. The speedups we have obtained are good, and the dynamic-compilation overhead is among the lowest of any dynamic-compilation system, typically 20–200 cycles per instruction generated on a Digital Alpha 21164. The majority of DyC's functionality has been used to dynamically compile an instruction-set simulator. Only three annotations were required, but a few other changes to the program had to be made due to DyC's lack of support for static global variables. This deficiency and DyC's rudimentary support for partially static data structures are the primary obstacles to making DyC easy to use.