This document provides an introduction to deep learning, including:
1) It explains the basic framework of machine learning as finding a function to map inputs to outputs, and introduces neural networks as a way to define a set of possible functions.
2) It outlines the three basic steps of deep learning: defining a set of functions, evaluating the goodness of each function, and selecting the best function.
3) It provides examples of neural network structures like fully connected feedforward networks, and applications like handwritten digit recognition.
3. WHAT IS LEARNING?
Learning is the one of the important psychological process determine the human behavior.
learning is the process of accumulation of knowledge, skills and attitudes. (KSA)
Learning may be through training, experience, reading, observation, discussions, electronic media
including internet, experimentation, facing the new challenges, travel and exploration etc. hence
training and development is more tools for learning.
Learning is not a one time activity or an annual, half yearly, quarterly or monthly activity. Learning
is a continues process
Learning is also an important source of motivation, Stimulation and job satisfaction
4.
5. Meaning of Learning
Learning is simply means that
permanent change in behavior
through education, training,
practice and experience
6.
7.
8. DIFFERENCE B/W TRAINING, DEVELOPMENT,
EDUCATION AND LEARNING
Training -Training is concerned with the teaching of specific, factual, narrow scoped
subject matter and skills. It is a formal classroom learning activities.
Development- Development is concerned with a broader subject matter of a
conceptual or theoretical nature and the development of personal attitudes. It
comprises all learning experiences, both on and off the job, including formal,
classroom training.
Education -Education, primarily, involves the presentation of material by the faculty
to students who are learning about the subject matter. The material being studied is
fundamentally well known material. Those activities known as teaching and training
are included in this category.
Learning- Learning is the process of acquiring knowledge or skill through study,
experience or teaching. It is a process that depends on experience and leads to long-
term changes in behavior potential. Behavior potential describes the possible
behavior of an individual in a given situation in order to achieve a goal.
10. GENERAL PRINCIPLES OF LEARNING
Readiness
Readiness implies a degree of concentration and eagerness.
Individuals learn best when they are physically, mentally, and
emotionally ready to learn, and do not learn well if they see no
reason for learning. Getting students ready to learn, creating
interest by showing the value of the subject matter, and
providing continuous mental or physical challenge, is usually
the instructor’s responsibility.
11. The principle of exercise states that those things most often
repeated are best remembered. It is the basis of drill
and practice. It has been proven that students learn best
and retain information longer when they have meaningful
practice and repetition. The key here is that the practice
must be meaningful. It is clear that practice leads to
improvement only when it is followed by positive feedback.
EXERCISE
12. EFFECT
The principle of effect is based on the emotional reaction of the student. It
has a direct relationship to motivation. The principle of effect is that learning
is strengthened when accompanied by a pleasant or satisfying feeling, and
that learning is weakened when associated with an unpleasant feeling. , every
learning experience should contain elements that leave the student with
some good feelings.
13. PRIMACY
The state of being first, often creates a strong,
almost unshakable, impression. Things learned
first create a strong impression in the mind that is
difficult to erase. For the instructor, this means
that what is taught must be right the first time. for
example, a student learns a faulty technique, the
instructor will have a difficult task correcting
bad habits and “re teaching” correct ones. The
student's first experience should be positive,
functional, and lay the foundation for all that is to
follow. What the student learns must be
procedurally correct and applied the very first
time.
14. RECENCY
The principle of recency states that things
most recently learned are
best remembered. Conversely, the further
a student is removed time-wise from a
new fact or understanding, the more
difficult it is to remember. For example, it
is fairly easy to recall a telephone number
dialed a few minutes ago, but it is usually
impossible to recall a new number dialed
last week.
15. INTENSITY
The principle of intensity implies that a
student will learn more from the real thing
than from a substitute. For example, a
student can get more understanding and
appreciation of a movie by watching it
than by reading the script. Likewise, a
student is likely to gain greater
understanding of tasks by performing
them rather than merely reading about
them.
16. REQUIREMENT
The law of requirement states that "we
must have something to obtain or do
something." It can be ability, skill,
instrument or anything that may help us
to learn or gain something. A starting
point or root is needed; for example, if
you want to draw a person, you need to
have the materials with which to draw,
and you must know how to draw a point,
a line, and a figure and so on until you
reach your goal, which is to draw a
person.
17. FREEDOM
The principle of freedom states that things freely learned are best learned.
Conversely, the further a student is coerced, the more difficult is for him to
learn, assimilate and implement what is learned. Compulsion and coercion
are antithetical to personal growth. The greater the freedom enjoyed by
individuals within a society, the greater the intellectual and moral
advancement enjoyed by society as a whole.
Since learning is an active process, students must have freedom: freedom of
choice, freedom of action, freedom to bear the results of action—these are
the three great freedoms that constitute personal responsibility. If no
freedom is granted, students may have little interest in learning.
24. 1. Deep Learning Tutorial 李宏毅 Hung-yi Lee
2. Deep learning attracts lots of attention. •I believe you haveseen lots of exciting results before. This talk focuses on thebasic techniques. Deep learning trends at Google. Source: SIGMOD/Jeff Dean
3. OutlineLecture IV: Next Wave Lecture III: Variants of Neural Network LectureII: Tips for Training Deep Neural Network LectureI: Introductionof Deep Learning
4. Lecture I: Introduction of Deep Learning
5. Outlineof Lecture I Introduction of Deep Learning Why Deep? “Hello World” for Deep Learning Let’s start with general machine learning.
6. MachineLearning ≈ Looking for a Function•Speech Recognition • Image Recognition • Playing Go • Dialogue System f f f f “Cat” “How are you” “5-5” “Hello”“Hi” (what theuser said) (system response) (next move)
7. Framework A set of function 21, ff 1f “cat” 1f “dog” 2f “money” 2f “snake” Model f “cat” Image Recognition:
8. Framework A set of function 21, ff f “cat” Image Recognition: Model Training Data Goodnessof function f Better! “monkey” “cat” “dog” function input:function output: Supervised Learning
9. Framework A set of function 21, ff f “cat” Image Recognition: Model Training Data Goodnessof function f “monkey” “cat” “dog” * f Pick the “Best” Function Using f “cat” Training Testing Step 1 Step 2 Step 3
10. Step 1: definea set of function Step 2:goodnessof functionStep3: pick thebest functionThreeSteps for Deep Learning Deep Learning is so simple ……
11. Step 1: definea set of function Step 2:goodnessof functionStep3: pick thebest functionThreeSteps for Deep Learning Deep Learning is so simple …… Neural Network
12. Human Brains
13. bwawawaz KKkk 11Neural Network z 1w kw Kw … 1a ka Ka b z bias a weights Neuron… …… A simplefunctionActivationfunction
14. Neural Network z bias Activation functionweightsNeuron1 -2 -1 1 2 -1 1 4 z z z e z 1 1 Sigmoid Function0.98
15. Neural Network z z z z Different connectionsleads to different networkstructureWeights and biases arenetwork parameters 𝜃 Each neurons can havedifferent values of weights and biases.
16. Fully Connect Feedforward Network z z z e z 1 1 Sigmoid Function 1 -1 1 -2 1 -1 1 0 4 -2 0.98 0.12
17. Fully Connect Feedforward Network 1 -2 1 -1 1 0 4 -2 0.98 0.12 2 -1 -1 -2 3 -1 4 -1 0.86 0.11 0.62 0.83 0 0 -2 2 1 -1
18. Fully Connect Feedforward Network 1 -2 1 -1 1 0 0.73 0.5 2 -1 -1 -2 3 -1 4 -1 0.72 0.12 0.51 0.85 0 0 -2 2 𝑓 0 0 = 0.51 0.85 Given parameters 𝜃, define a function 𝑓 1 −1 = 0.62 0.83 0 0 This is a function. Inputvector, output vector Given network structure, definea function set
19. Output LayerHidden Layers InputLayer Fully Connect Feedforward Network InputOutput1x 2x Layer 1 …… Nx …… Layer 2 …… Layer L …… …… …… …… …… y1 y2 yM Deep means many hidden layers neuron
20. Output Layer (Option) • Softmax layer as theoutputlayer Ordinary Layer 11 zy 22 zy 33 zy 1z2z 3z In general, the outputof network can beany value. May not beeasy to interpret
21. Output Layer (Option) • Softmax layer as theoutputlayer 1z 2z 3z Softmax Layer e e e 1z e 2z e 3z e 3 1 1 1 j zz j eey 3 1j z j e 3 -3 1 2.7 20 0.05 0.88 0.12 ≈0 Probability: 1 > 𝑦𝑖 > 0 𝑖 𝑦𝑖 = 1 3 1 2 2 j zz j eey 3 1 3 3 j zz j eey
22. Example Application InputOutput 16 x 16 = 256 1x 2x 256x …… Ink → 1 No ink → 0 …… y1 y2 y10 Each dimension represents theconfidenceof a digit. is 1 is 2 is 0 …… 0.1 0.7 0.2 The image is “2”
23. Example Application •Handwriting Digit Recognition Machine “2” 1x 2x 256x …… …… y1 y2 y10 is 1 is 2 is 0 …… What is needed is a function …… Input:256-dimvector output: 10-dim vector Neural Network
24. Output LayerHidden Layers InputLayer Example Application InputOutput 1x 2x Layer 1 …… Nx …… Layer 2 …… Layer L …… …… …… …… “2” …… y1 y2 y10 is 1 is 2 is 0 …… A function set containing thecandidates for Handwriting Digit Recognition You need to decidethenetwork structureto let a good functionin your function set.
25. FAQ • Q: How many layers? How many neuronsfor each layer? • Q: Can thestructurebeautomatically determined? Trial and Error Intuition+