Training  ConvNets  with  Torch   Soumith  Chintala   Next.ml   17th  January,  2014  

Tutorial  Structure   Hour  1   •  AcBon  RecogniBon  from  Videos:  

–  Discuss  task  at  hand   –  End-­‐to-­‐end  rule-­‐based  methods   –  Hand-­‐engineered  features  +  SVM   –  Neural  Nets  (and  discuss  discovering  graph  structure   automaBcally)   –  ConvNets        

  •  Notebook  Setup  

Tutorial  Structure   Hour  2   •  Torch  Basics   –  –  –  –  – 

•  • 

Basic  visualizaBon  in  iTorch   Neural  Networks   –  –  –  –  –  – 

• 

What  is  torch?       Lua  and  Syntax   Tensors,  types,  storages   Slicing  and  selecBng   views  vs  copies  

Containers  and  Modules   Examples:   Overfeat   Alexnet   IncepBon   LSTM  

OpBm  package  

Tutorial  Structure   •  Hour  2/3   •  Coding  an  acBon  recogniBon  system   –  load  and  normalize  data   –  create  neural  network   –  train  neural  network   –  test/validate  neural  network  

 

Task  at  Hand  

First  a\empt   Pure  rule-­‐based  

Second  A\empt     •  Hand-­‐designed  features  

Third  A\empt   •  ConvoluBon  Networks  

Fourth  A\empt   •  Fully  connected  nets  

Neural  net  in  detail   •  Show  a  typical  two-­‐layer  network  

•  ……………  

What  is  Torch?   •  Not  a  language.  Based  on  Lua   •  Not  just  a  Tensor  library   •  Ecosystem  and  community  

Lua  and  Syntax  

Tensors   •  nDimensional  arrays  –  torch.Tensor   •  Separated  from  Storages   •  Can  be  sliced,  re-­‐viewed,  selected  etc.  without   a  memory  copy  where  possible  

Basic  VisualizaBon  

Neural  Networks  

Neural  Networks  

Neural  Networks  

OpBmizaBon  

Training ConvNets with Torch - GitHub

Jan 17, 2014 - ... features + SVM. – Neural Nets (and discuss discovering graph structure automařcally). – ConvNets. • Notebook Setup ... Page 9 ...

1MB Sizes 0 Downloads 251 Views

Recommend Documents

Test Harness Training - GitHub
Jun 29, 2016 - operational Analytic Development Test. Harness is now ..... Open the Data Modeling Manager Tool. openECA Client app starts the web tool ...

Research Data Management Training - GitHub
Overview. Research Data management Training Working Group: Approach and. Methodology ... CC Australia ported licence) licence. ... http://www.griffith.edu.au/__data/assets/pdf_file/0009/528993/Best_Practice_Guidelines.pdf. University of ...

torch-ebook.pdf
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. torch-ebook.pdf.

Android Training with Placement
Why Android is different (and important). Android Stack 30 min. •. Overview of the stack. •. Linux kernel. •. Native libraries. •. Dalvik. •. App framework. •. Apps. SDK Overview 1hr. •. Platforms. •. Tools. •. Versions. Hello World

Android Training with Placement
Simple media playback. •. Supported video formats. •. Simple video playback. SQL Database 4 hrs. •. Introducing SQLite. •. SQLite Open Helper and creating a database. •. Opening and closing a database. •. Working with cursors Inserts, upd

Android Training with Placement
Simple media playback. •. Supported video formats. •. Simple video playback. SQL Database 4 hrs. •. Introducing SQLite. •. SQLite Open Helper and creating a database. •. Opening and closing a database. •. Working with cursors Inserts, upd

Android Training with Placement
Simple media playback. •. Supported video formats. •. Simple video playback. SQL Database 4 hrs. •. Introducing SQLite. •. SQLiteOpenHelper and creating a database. •. Opening and closing a database. •. Working with cursors Inserts, updat

CP2K with LIBXSMM - GitHub
make ARCH=Linux-x86-64-intel VERSION=psmp AVX=2. To target for instance “Knights ... //manual.cp2k.org/trunk/CP2K_INPUT/GLOBAL/DBCSR.html).

Java with Generators - GitHub
processes the control flow graph and transforms it into a state machine. This is required because we can then create states function SCOPEMANGLE(node).

Past Torch Winners 2016.pdf
RestorAid. Rumpke Consolidated Companies. Total Quality Logistics, Inc. 2004 ... Page 1 of 1. Main menu. Displaying Past Torch Winners 2016.pdf. Page 1 of 1.

OpenBMS connection with CAN - GitHub
Arduino with BMS- and CAN-bus shield as BMS a master. - LTC6802-2 or LTC6803-2 based boards as cell-level boards. - CAN controlled Eltek Valere as a ...

Better performance with WebWorkers - GitHub
Chrome52 on this Laptop. » ~14kbyte. String => 133ms ... 3-4 Seks processing time on samsung galaxy S5 with crosswalk to finish the transition with ... Page 17 ...

with ZeroMQ and gevent - GitHub
Normally, the networking of distributed systems is ... Service Oriented .... while True: msg = socket.recv() print "Received", msg socket.send(msg). 1. 2. 3. 4. 5. 6. 7.

Getting Started with CodeXL - GitHub
10. Source Code View . ..... APU, a recent version of Radeon Software, and the OpenCL APP SDK. This document describes ...... lel_Processing_OpenCL_Programming_Guide-rev-2.7.pdf. For GPU ... trademarks of their respective companies.

Getting Started with Go - GitHub
Jul 23, 2015 - The majority of my experience is in PHP. I ventured into Ruby, ... Compiled, Statically Typed, Concurrent, Imperative language. Originally ...

ConvNets-Based Action Recognition from Depth Maps ...
Fig.2 Kinect Sensors source from Apple. Fig.3 Deep Learning source from VLab MIT. REFERENCE. 1. B. R. Abidi, Y. Zheng, A. V. Gribok, and M. A. Abidi, ...

Getting Acquainted with R - GitHub
In this case help.search(log) returns all the functions with the string 'log' in them. ... R environment your 'working directory' (i.e. the directory on your computer's file ... Later in the course we'll discuss some ways of implementing sanity check

Examples with importance weights - GitHub
Page 3 ... Learning with importance weights y. wT t x wT t+1x s(h)||x||2 ... ∣p=(wt−s(h)x)Tx s (h) = η. ∂l(p,y). ∂p. ∣. ∣. ∣. ∣p=(wt−s(h)x)Tx. Finally s(0) = 0 ...

Digital Design with Chisel - GitHub
Dec 27, 2017 - This lecture notes (to become a book) are an introduction into hardware design with the focus on using the hardware construction language Chisel. The approach of this book is to present small to medium sized typical hardware components

Deep Learning with H2O.pdf - GitHub
best-in-class algorithms such as Random Forest, Gradient Boosting and Deep Learning at scale. .... elegant web interface or fully scriptable R API from H2O CRAN package. · grid search for .... takes to cut the learning rate in half (e.g., 10−6 mea

10 Watt NDT Torch Spec.pdf
365nm UV LED Inspection Torch. Whoops! There was a problem loading this page. 10 Watt NDT Torch Spec.pdf. 10 Watt NDT Torch Spec.pdf. Open. Extract.

pdf viewer blackberry torch
File: Pdf viewer blackberry torch. Download now. Click here if your download doesn't start automatically. Page 1 of 1. pdf viewer blackberry torch. pdf viewer ...