Вы находитесь на странице: 1из 15

COURSE OVERVIEW

Trn Trung Kin


ttkien@fit.hcmus.edu.vn

Video about machine learning class in Coursera

9/12/2014

Machine Learning

Not machine learning

We do NOT know f,
but we have DATA
which can be used to approximate f
Input Output

9/12/2014

prime

We know f

Conditions to apply machine learning

A pattern exists.
We cannot pin it down mathematically.
We have data on it.

9/12/2014

A lot of super cool apps


Machine learning is used in situations where we dont have
an analytic solution, but we have data which can be used to
construct an empirical solution It can be used in a lot of
fields:
AI: computer vision, audio processing, speech processing,

NLP, ...
Medicine.
Economics.
9/12/2014

MNIST dataset

Better than
human
9/12/2014

The ILSVRC-2012 competition on ImageNet

The dataset has 1.2 million


high-resolution training images.
The classification task:
Get the correct class in
your top 5 bets. There are
1000 classes.
The localization task:
For each bet, put a box
around the object. Your box
must have at least 50%
overlap with the correct box.

Some of the best existing


computer vision methods were
tried on this dataset by leading
computer vision groups from
Oxford, INRIA, XRCE,
Computer vision systems
use complicated multistage systems.
The early stages are
typically hand-tuned by
optimizing a few
parameters.

Hintons slide

Examples from the test set (with the networks


guesses)

Hintons slide

University of Toronto (Alex Krizhevsky)

16.4%

34.1%

Error rates on the ILSVRC-2012 competition


classification

University of Tokyo
Oxford University Computer Vision Group
INRIA (French national research institute in CS) +
XRCE (Xerox Research Center Europe)
University of Amsterdam

Hintons slide

26.1%
26.9%
27.0%

29.5%

classification
&localization

53.6%
50.0%

Clarifai (clarifai.com)
See how well computers can understand images.

9/12/2014

10

Machine learning is hot nowadays


If you want to be in academia, ML is a good place for you.
Deep Learning.

If you want to be in industry, ML is a good place for you too.


In big IT companies like Google, Facebook, ..., ML is one of the most
important skills.
In Viet Nam, recently some ML companies have emerged .

9/12/2014

11

The way we do in this course


We will follow the excellent online course taught by Caltech Professor

Yaser Abu-Mostafa. The lectures go as follows:


What is machine learning?
Can a machine learn?
How to do it?

How to do it well?
Take-home lessions.

There are 18 lectures. With 14 weeks (skip this week), we will go

through the first 14 lectures; the rest will be left as your homework.

9/12/2014

12

The way we do in this course


Before class, you will watch the video lecture I
require, and post to the forum at least one question.
In class, I will review the lecture based on your
questions; we will discuss to each other.
After class, you will pick at least one question and
post your answer to the forum.

9/12/2014

13

The way we do in this course


We will have 1 exercise / 2 video lectures.
Programming language: Octave/Matlab. It allows us to
implement ML algorithms very quickly.
The code should be written and commented clearly.

No exams .
Grading:
Exercises: 80%.

Forum Q&A: 30%.


9/12/2014

14

The way we do in this course


DONT CHEAT.
You guys are allowed to discuss to each other as well as
consult any source, but NOT allowed to copy from others or
from any source.
If you violate this rule 1 time, you will get 0 score for the
corresponding exercise.
If you violate this rule > 1 time, you will get 0 score for the

course.
9/12/2014

15

Вам также может понравиться