Introduction to
Machine Learning
Odyssée Merveille and Emmanuel Roux CREATIS, Lyon
-
A short historical background
-
Supervised Learning
-
Unsupervised Learning
-
Conclusion
Content
A short historical background
Machine Learning
Deep Learning
Artificial Intelligence
Inspired (and simplified) from the deeplearningbook.org
(I. Goodfellow and Y. Bengio, A. Courville, 2016)
A short historical background
Machine Learning
Deep Learning
Artificial Intelligence
Inspired from Sebastian Raschka's deep-learning course
A short historical background
Artificial Intelligence
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
A short historical background
Artificial Intelligence
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
DEDUCTIVE
rule-based
no need of examples
INDUCTIVE
example based
adaptation
Symbolic AI
connectionism
A short historical background
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
data
outputs
Symbolic AI
program
A short historical background
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
data
expected outputs
connectionism
program
A short historical background
Artificial Intelligence
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
DEDUCTIVE
rule-based
no need of examples
INDUCTIVE
example based
adaptation
Symbolic AI
connexionism
A short historical background
Cybernetics (40’s to 60’s)
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
Symbolic AI
connexionism
Perceptron (Rosenblatt)
ADALINE (Widrow & Hoff)
(W. Ross Ashby)
source wikipedia
https://isl.stanford.edu/~widrow/papers/t1960anadaptive.pdf
A short historical background
Symbolic Artificial Intelligence (60’s to 80’s)
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
Symbolic AI
connexionism
MYCIN (Shortliffe): medical diagnoses (bacteria identification)
GUIDON (Clancey): teaching medical diagnostic strategy
CADUCEUS (Pople): internal medicine expert system
A short historical background
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
publication trends timeline
negative feedback
logic-based rules
backprop
McCulloch, Pitts, Hebb
Samuel, Ashby, Rosenblatt
McCarthy, Minsky, Papert, Simon,
Newell
Hinton, LeCun,
Breiman, Rumelhart
A short historical background
Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)
ideas trends timeline
negative feedback
logic-based rules
backprop
CIFAR-NCAP
Datasets++
GPU++
"Deepstributed"
Perceptron (Rosenblatt)
SVM (Cortes, Vapnik)
ADALINE (Widrow & Hoff)
Homoestat (Ashby)
MYCIN (Shortliffe)
GUIDON (Clancey)
INTERNIST (Myers)
Perceptrons (Minsky, 1969)
backpropagation
(Rumelhart, )
Deep Belief Nets (Hinton)
Machine Learning
A computer program is said to learn [...] if its performance at tasks in T, as measured by a performance indicator P, improves with experience E.
reformulation of Tom Mitchel, 1978
A short historical background
Machine learning is the field of study that gives computers the ability to learn
without being explicitly programmed.
Arthur L. Samuel, AI pionneer, 1959
A computer program is said to learn [...] if its performance at tasks in T, as measured by a performance indicator P, improves with experience E.
reformulation of Tom Mitchel, 1978
A short historical background
Machine Learning
Machine learning is the field of study that gives computers the ability to learn
without being explicitly programmed.
Arthur L. Samuel, AI pionneer, 1959
A computer program is said to learn [...] if its performance at tasks in T, as measured by a performance indicator P, improves with experience E.
Tom Mitchel, 1978 (tweaked citation)
A short historical background
A short historical background
Images from PhD student
Ludmilla Penarrubia
Task
Experiment
Performance
Learn
A short historical background
supervised learning
Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.
A short historical background
supervised learning
Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.
A short historical background
supervised learning
Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.
A short historical background
supervised learning
Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.
A short historical background
supervised learning
unsupervised learning
- detection (lesions)
- classification (benign/malign)
- segmentation (organs)
- prediction (prognostic)
- ...
- detection (lesions)
- classification (benign/malign)
- segmentation (organs)
- prediction (prognostic)
- ...
- clustering
- dimension reduction
- representation
- density estimation
- ...
A short historical background
supervised learning
unsupervised learning
Image from PhD student
Yamil Vindas
A short historical background
supervised learning
unsupervised learning
reinforcement
learning
self-supervised
learning
Transfert
learning
Domain Adaptation
supervised learning
unsupervised learning
semi-supervised
learning
reinforcement
learning
self-supervised
learning
Transfert
learning
Domain Adaptation
Weakly-supervised
A short historical background
supervised learning
supervised learning
unsupervised learning
unsupervised learning
Conclusion
supervised learning
unsupervised learning
semi-supervised
learning
Weakly-supervised
Talk by Ismail Ben Ayed and Jose Dolz
Friday April 23
4 pm - Paris time
Conclusion
Symbolic AI
connectionism
#responsibleAI (biases, ethics)
« a priori » within learning
Hands-on session 2.x
explanable AI (xAI)
Talk by Narine Kokhlikyan
Tuesday April 20
4.20 pm - Paris time
Conclusion
A short historical background
supervised learning
- The Digital Database for Screening Mammography, Michael Heath, Kevin Bowyer, Daniel Kopans, Richard Moore and W. Philip Kegelmeyer, in Proceedings of the Fifth International Workshop on Digital Mammography, M.J. Yaffe, ed., 212-218, Medical Physics Publishing, 2001. ISBN 1-930524-00-5.
- Task
- Experiment
- Performance
- Learn
data
model
training method
loss function
A short historical background
replace the process_fcuntion with model. loss_function... (or criterion...) etc.
source https://pytorch.org/ignite/concepts.html