ApDeepSense: Deep Learning Uncertainty Estimation Without the Pain for IoT Applications

Author:

Shuochao Yao, Yiran Zhao, Huajie Shao, Chao Zhang, Aston Zhang, Dongxin Liu, Shengzhong Liu, Lu Su, Tarek F. Abdelzaher

Intro:

As a result of the recent advancement of deep learning applications, growing attention from the IoT community has been attracted to this technique. Different from other applications, the one with IoT areas require the learning method provide outputs as well as the uncertainty estimates of the outputs. The current state-of-art solution involves intensive sampling procedure, or even re-train the network, which is too expensive for many small applications. This paper presents ApDeepSense; a method saves around 90% of energy and time consumptions while producing more accurate uncertainty estimates.

Methodology:

The current state-of-art used dropout training to estimate the uncertainty and proved that the equivalence between dropout training and the approximate inference in a deep Gaussian process. This paper, however, showed that the internal-layer representation of deep neural networks could be represented by an internal-layer representation of multivariate Gaussian distributions. And by applying the matrix multiplication and activation functions to the probabilistic distribution input, we can get not only the expected value of output but also the probability distribution of that output. 

Authors also tried to define special operations to deal the probabilistic distribution input, which includes using element-wise multiplication to represent the matrix form of expected value and variance of output and use piece-wise linear functions as the closed-form solution for non-linear activation function.

Critique:

1. This paper's result is amazing, it provides good result while using less time and energy. You get the output for the expected value and the uncertainty of the output from the variance. The basic idea here is you can use the mean and variance from the previous layer to get the mean and variance for the next layer. I think the only problem with this approach is that it assumes the input always has a normal distribution, it will be problematic when the input has some other underlying distribution.

2. One limitation of this paper is that it only considered fully-connected layers, limited the type of network this paper could apply. But I don't think this would be really big problem.

That's all I have for this paper. I cannot find any other flaws I can attack. Even the proof is good, it's right amount of math that you don't spend hours to go through yet is long enough that you feel you are smart.

--Yifan Hu

Comments

  1. jon_weissmanApril 16, 2019 at 6:47 AM

    I'm impressed that you worked through some of the mathematics! The problem of uncertainty estimation appears to be critical in human-in-the-loop applications such as many IoT scenarios. Assumption of Gaussian/Normal seem fine given the physical nature of the data in the IoT scenarios.

    ReplyDelete

Post a Comment

Popular posts from this blog

A Machine Learning Approach to Live Migration Modeling

Authors:  Changyeon Jo, Youngsu Cho, Bernhard Egger Seoul National University Summary:  This paper proposed a machine learning approach to predict the system metrics for different VM migration algorithms. These system metrics are affected by migration algorithm, host machine status and user applications and feasible for machine learning approach. The prediction model takes 21 system/application input features (including which algorithm), and generates 6 metrics as the output. As a result, it can accurately estimate each metrics under different application, VM and algorithms. One of the application, the paper provide a semi-automated algorithm selection based on this machine learning model and user-provided SLA (service level agreements) constraints. Questions about Machine Learning:  Why machine learning is applied: The combination of algorithm, VM status and current running applications status, is a huge searching space. Machine learning features: which algor...
Read more

StormDroid: A Streaminglized Machine Learning-Based System for Detecting Android Malware

Introduction: The paper approaches the problem of Malware Detection using machine learning approach. In this paper, the authors propose considering new type of features along with two traditionally used set of features for developing techniques for malware detection. Along with this proposal, the paper makes other contributions as well. The authors provide a new framework which is able to process information from large data sets as streams, and efficiently perform malware detection. Motivation: Android is an open platform for development. It has certainly made easy for everyone to develop and publish apps.  The number of apps published on Google store is in millions. However, it has also made easier for hackers to reach out to a larger population for malicious attacks. Additionally, with surge in the number of attacks on devices, more data is available for analysis which can be used by the machine learning models to identify malicious apps. This is motivation to use machine ...
Read more

Coordinated Management of Multiple Interacting Resources in Chip Multiprocessors: A Machine Learning Approach

Authors: Ramazan Bitirgen, Engin Ipek, Jose F. Martinez Motivation: Sharing system resources efficiently on chip multiprocessors (CMPs) platforms is the key to obtain high utilization and enforce system-level performance objectives. Unrestricted sharing of microarchitectural resources will result in destructive interference and unacceptable performance. However, the works focus on coordinated management of multiple interacting resources on CMPs are very scarce , even though several proposals focusing on addressing the management of a single microarchitectural resource exist. Aiming to fill this gap, this paper proposes a framework with the machine learning technology to manage multiple shared CMP resources in a coordinated fashion to enforce higher-level performance objectives. This novel mechanism consists of per-application hardware performance models and a global resource manager. Positive Points: 1. The global resource manager is able to aggregate each app...
Read more