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

Machine Learning

No description
by

Xianda Sun

on 23 October 2012

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Machine Learning

A brief Intro to Machine Learning How Do Machines Learn? Before We Start... some examples Google Face
Recognition Machine Learning Image Search Apple Siri Facebook Definition Friends Recommendation Provide you with similar images Know your request by listening to you Shared friends?
Geographically close?
Common experience? How does it managed to do? In the past, computer can
do things ONLY if you told it how to. To program on Eniac In assembly lang:

10 INT 21h
20 LOAD r1, ah
20 LOAD r2, 1
30 ADD r1, r2 In X86, this is for reading a number, adding 1 to it and storing in register r1. We want the machine to do it by itself Machine Learning: Field of study that gives computers the ability to learn without being explicitly programmed. --Arthur Samuel Developer of the Checker program, world's first self-learning program A typical Model of Generalization Data The model built
on these Data New Data What user want How? Firstly, An Example Titanic Survivors From Wikipedia, I take it because it's reasonalbe and not so boring. At the last time of Titanic, only a small number of
people can escape the doom of death:

Given the condition of a people, can a machine tell if he/she survived? With the data we build a decision tree Finish when no attributes left Choose the attribute that best classify the sample Machine
Learning If there are more attributes left, repeat the former step on each subset of Data Firstly, we need data The condition includes:
1. Age
2. Sex
3. Number of siblings Part of the list of passengers, and their result. Decision
Tree
Model Data Cluster Data
Using 3 attributes Age Sex Siblings <9.5 >9.5 M F >2.5 <2.5 Count the survival
percentage
of each group 73% survived Build the root of the Dec Tree Using this attribute Entropy Sex M F Survived Note that machines are not always right,
they give the most likely answer. Note that under M and F it is not finished We practise the same step on:

{s in Data | s.sex = M}
and
{s in Data | s.sex = F} Build a dec-tree some conditions are pruned
Use this tree to predict the unknown people Types of Machine Learning Reinforcement Learning Unsupervised Learning Semi-supervised Learning Machine Learning Supervised Learning Training data:
inputs and correct answers Only inputs Training data is
a combination. Learn Action
with feedback
from environment. Learning Algorithms Decision Tree Friends Suggestion (Skipped) Example of Unsupervised Learning What do you share in common with your friends? Same hobby, same city, same social circle... Who will become your friends? Same hobby, same city, same social circle... Cluster People! SOCIAL NETWORKS have information about your age, schools, hobbies, social circle, geo-location... (so private! :) Imagine the world is an N-dimension space, and every people are points in this space.
The axes are age, geo-locations, hobbies and so on...
We can calculate distance between every two. You One of
your friends distance We use cluster algorithms - kmeans
to cluster everybody in the world~ KMEANS A machine learning model Suggest those who are in the same cluster to you. Impact - Can machines think? :) Machines become more sophisticated, liberate the programmers. :( Machine Learning provides results with uncertainty.
And can be low-efficient. Can Machines Think? From the previous examples, machine learning is nothing more than a bunch of algorithms. Nothing mysterious actually! However... Some argues that it can think as long as it behaves like a person. Turing Chinese Room Room A person doesn't understand or speak Chinese But Has All the Rules of Answering Questions in Chinese: e.g. People outside:

He knows Chinese! What do you think of it? Thank you! BY: Xianda Sun Service condition1 cond cond res res res res res res input function f(input) -> output f(inter-res1, inter-res2, inter-res3) -> output f(o1, o2, o3, o4) e.g. Input Method in___ ... What is the most likely
character following 'n' ...
put
stant
... Divide the nodes
in three clusters Training
Data M1 M2 M3 Validating
Data Get the correct percentage
of every machine System answers questions by voting,
better machine has more priority The rule is like:
if you are asked 'How are you?',
you shall answer 'Fine, thank you.' e.g. Used for predicting a Y/N answer e.g. Used for pattern recognition e.g. Used for predicting on current conditions e.g. Used for similarity detection More solid and precise than naive algs Artificial Neural Network Bayesian Clustering Ensemble Learning
Full transcript