Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Neural Networks

Ye
by

Pine Pineh

on 20 March 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Neural Networks

or: Rise of the Machines The Basics Artificial Neural Networks
Google X Labs

Self driving vehicles
Augmented reality glasses
Space elevators How can a machine learn?
System of interconnected artificial neurons
Adapt to and seek patterns within input data AKA Google Brain What's with the name? There's a reason it's a network

Traditionally refers to the mammalian brain's framework

Biological neurons connected by synapses

Activated by recognition of known stimuli

Specific groups of neurons perform the task-specific functions they have adapted to

Frequency of activation strengthens neural connections - recognition

Google's artificial neural network functions similarly

Operates by recognizing stimuli, creating unique categories, and strengthening recognition
Many current recognition systems rely on

Objects defined only by low level features SOME NUMBERS Google What if we could scale it up to the magnitude of a brain? What's the big deal? Deep Learning Surface Learning Artificial Neural Network One of Google's Neuron's represents a completely unique high level feature Image Map - Set of Pixels representing an image

Compare to and categorize unlabeled data

Capable of Recognition - Activation Values The Problem amounts of unlabeled data exist Huge It is useless Harnessing the potential and applying it Brain 3 Million Neurons 1 Billion Connections 100 Billion Neurons 100 Trillion Synapses neural Cat Neuron Face Neuron Human Body Neuron ? Neuron Google Brain: 1000 machines
16000 cores What is Recognition? Network can recognize invariance in an input image

Categorize similar data into unique but unlabeled groups The machine only knows that different entities exist - but not what they are What makes that exciting? A human child learns in much the same way - with some guidance Google's neural network does not have the advantage of nature on its side And it doesn't require any amount of instruction That means Google Brain is capable of conceptualization the idea of a from only raw data Create cat cat face body No instruction other than an algorithm and
unlabeled, meaningless data Causes inaccuracy in current
recognition systems Google's artificial neurons respond to extremely specific high level features What does a neuron really do? SPARSE CODING ImageNet Dataset
22k Categories, 14m Images
15.8 9.3 "Cat Detector" Computes extremely complex function of the input
Function creates useful representation of the input Apply It Data Labeling 10m YouTube Images 200x200 3 Days - Neurons Become Selective Financial Potential REALLY GOOD Google X Labs ---> Search & Services Division supervised pretraining The Programmer Chooses POSITRONIC BRAIN Character Recognition Stock Prediction Applications (Better than a Human) PREDICTIONS Sociology Speech Recognition
Full transcript