DeepSense: a unified deep learning framework for time-series mobile sensing data processing

May 17, 2017 ~ Adrian Colyer ~ 5 Comments

DeepSense: a unified deep learning framework for time-series mobile sensing data processing Yao et al., WWW'17 DeepSense is a deep learning framework that runs on mobile devices, and can be used for regression and classification tasks based on data coming from mobile sensors (e.g., motion sensors). An example of a classification task is heterogeneous human ... Continue Reading

Usage patterns and the economics of the public cloud

May 16, 2017 ~ Adrian Colyer ~ 6 Comments

Usage patterns and the economics of the public cloud Kilcioglu et al., WWW'17 Illustrating the huge diversity of topics covered at WWW, following yesterday's look at recovering mobile user trajectories from aggregate data, today's choice studies usage variation and pricing models in the public cloud. The basis for the study is data from 'a major ... Continue Reading

Trajectory recovery from Ash: User privacy is NOT preserved in aggregated mobility data

May 15, 2017 ~ Adrian Colyer ~ 11 Comments

Trajectory recovery from Ash: User privacy is NOT preserved in aggregated mobility data Xu et al., WWW'17 Borrowing a little from Simon Wardley's marvellous Enterprise IT Adoption Cycle, here's roughly how my understanding progressed as I read through this paper: Huh? What? How? Nooooo, Oh No, Oh s*@\#! Xu et al. show us that even ... Continue Reading

Learning to act by predicting the future

May 12, 2017November 11, 2019 ~ Adrian Colyer ~ 5 Comments

Learning to act by predicting the future Dosovitskiy & Koltun, ICLR'17 The Direct Future Prediction (DFP) model won the 'Full Deathmatch' track of the Visual Doom AI Competion in 2016. The competition pits agents against each other, with their performance measured by how many 'frags' they get. (A frag is a kill, apparently - not ... Continue Reading

Understanding deep learning requires re-thinking generalization

May 11, 2017November 11, 2019 ~ Adrian Colyer ~ 33 Comments

Understanding deep learning requires re-thinking generalization Zhang et al., ICLR'17 This paper has a wonderful combination of properties: the results are easy to understand, somewhat surprising, and then leave you pondering over what it all might mean for a long while afterwards! The question the authors set out to answer was this: What is it ... Continue Reading

Neural architecture search with reinforcement learning

May 10, 2017November 11, 2019 ~ Adrian Colyer ~ 7 Comments

Neural architecture search with reinforcement learning Zoph & Le, ICLR'17 Earlier this year we looked at 'Large scale evolution of image classifiers' which used an evolutionary algorithm to guide a search for the best network architectures. In today's paper, Zoph & Le also demonstrate that learning network architectures (and also in their case recurrent cell ... Continue Reading

Semi-supervised knowledge transfer for deep learning from private training data

May 9, 2017November 11, 2019 ~ Adrian Colyer ~ 1 Comment

Semi-supervised knowledge transfer for deep learning from private training data Papernot et al., ICLR'17 How can you build deep learning models that are trained on sensitive data (e.g., concerning individuals), and be confident to deploy those models in the wild knowing that they won't leak any information about the individuals in the training set? As ... Continue Reading

End-to-end optimized image compression

May 8, 2017November 11, 2019 ~ Adrian Colyer ~ Leave a comment

End-to-end optimized image compression Ballé et al., ICLR'17 We'll be looking at some of the papers from ICLR'17 this week, starting with "End-to-end optimized image compression" which makes a nice comparison with the Dropbox JPEG compression paper that we started with last week. Of course you won't be surprised to find out that the authors ... Continue Reading

Efficient memory disaggregation with Infiniswap

May 5, 2017November 11, 2019 ~ Adrian Colyer ~ 3 Comments

Efficient memory disaggregation with Infiniswap Gu et al., NSDI '17 If we move performance numbers onto a human scale (let 1ns of processor time = 1 second of human time) then it's easier to get an intuition - for me at least - of the relative cost of different operations. In this world, it takes ... Continue Reading

CherryPick: Adaptively unearthing the best cloud configurations for big data analytics

May 4, 2017November 11, 2019 ~ Adrian Colyer ~ 14 Comments

CherryPick: Adaptively unearthing the best cloud configurations for big data analytics Alipourfard et al., NSDI'17 For big data analytics jobs, especially recurring jobs, finding a good cloud configuration (number and type of machines, CPU, memory ,disk and network options) can make a big different to overall cost and runtimes. Likewise, a poor choice can seriously ... Continue Reading

the morning paper

a random walk through Computer Science research, by Adrian Colyer
Made delightfully fast by strattic

Month: May 2017