SLIDE 1
WHO IS TEUBI?
listening to
NUSSSCHALE PODCAST
introduction to WHO IS TEUBI? listening to NUSSSCHALE PODCAST - - PowerPoint PPT Presentation
introduction to WHO IS TEUBI? listening to NUSSSCHALE PODCAST - - PowerPoint PPT Presentation
WHO IS TEUBI? introduction to WHO IS TEUBI? listening to NUSSSCHALE PODCAST WHO IS TEUBI? graebel et al. evaluating out-of-the-box methods for the classification of hematopoietic cells in images of stained bone marrow ALGORITHMS THEN
SLIDE 2
SLIDE 3
WHO IS TEUBI?
graebel et al. evaluating out-of-the-box methods for the classification of hematopoietic cells in images of stained bone marrow SLIDE 4
ALGORITHMS THEN & NOW
some input some output algorithm
> do this > do that > if this > then that > else > this
SLIDE 5
ALGORITHMS THEN & NOW
some input some output machine learning
SLIDE 6
ALGORITHMS THEN & NOW
some input some output machine learning size speed
SLIDE 7
OVERFITTING UNDERFITTING
SLIDE 8
OVERFITTING UNDERFITTING
SLIDE 9
OVERFITTING UNDERFITTING
SLIDE 10
NEURAL NETWORKS
SLIDE 11
NEURAL NETWORKS
i1 i3 i2
- w1
w2 w3
- = f (Σ iw)
f(x) x
SLIDE 12
NEURAL NETWORKS
speed 10 size 1 1
- 1
speed + size > 10 speed + size < 10 1
f(x) x
1
SLIDE 13
NEURAL NETWORKS
INPUT OUTPUT OUTPUT OUTPUT computed expected 1 1 1 1
raise weights! lower weights!
|| ocomputed - oexpected ||
λ,
SLIDE 14
NEURAL NETWORKS
LAYER hidden LAYER hidden LAYER
- utput
LAYER input weights LAYERS number of NEURONS number of FUNCTION activation RATE learning
SLIDE 15
NEURAL NETWORKS
CLASSIFICATION TASKS ANIMAL rabbit | unicorn TEXT spam | not spam IMAGE cat | dog PATIENT healthy | ill size speed colour
SLIDE 16
NEURAL NETWORKS
CLASSIFICATION TASKS ANIMAL rabbit | unicorn TEXT spam | not spam IMAGE cat | dog PATIENT healthy | ill sender recipient hyperlinks? certain words?
SLIDE 17
NEURAL NETWORKS
CLASSIFICATION TASKS ANIMAL rabbit | unicorn TEXT spam | not spam IMAGE cat | dog PATIENT healthy | ill medical history temperature certain symptoms
SLIDE 18
NEURAL NETWORKS
CLASSIFICATION TASKS ANIMAL rabbit | unicorn TEXT spam | not spam IMAGE cat | dog PATIENT healthy | ill pixel values shape? texture?
SLIDE 19
IMAGE PROCESSING
SLIDE 20
IMAGE PROCESSING
SLIDE 21
CONVOLUTIONAL LAYER
SLIDE 22
CONVOLUTIONAL LAYER
SLIDE 23
DEEP LEARNING
SLIDE 24
DEEP CONVOLUTIONAL NET
SLIDE 25
SEGMENTATION NETWORK
SLIDE 26
SEGMENTATION NETWORK
SLIDE 27
SUPERVISED TRAINING
SLIDE 28
NETWORK ARCHITECTURES
FULLY CONVOLUTIONAL DECODER FULLY CONNECTED NETWORK FULLY CONVOLUTIONAL ENCODER MAX POOLING LAYER DROPOUT LAYER
SLIDE 29
GENERATIVE ADVERSARIAL NETS
SLIDE 30
GENERATIVE ADVERSARIAL NETS
Discriminator
Generator
SLIDE 31
GENERATIVE ADVERSARIAL NETS
Discriminator Generator
1 5 6 2 3 7 8 4 9
SLIDE 32
CYCLE GAN
Discriminator Generator Discriminator Generator
SLIDE 33
ADVERSARIAL EXAMPLES
Network Adversarial Network
SLIDE 34
WORD LANGUAGE MODEL
IN DIESER EPISODE: FÄDEN!
SLIDE 35
WORD LANGUAGE MODEL
IN DIESER EPISODE: FÄDEN! JERUSALEM! die katzen! stellt ihr jetzt über die mathematik. nichtsdestotrotz ist wasserstoff-atome ein lateinische neutron um ein etwas, sie kein schale, umrechnet, hotel, oder zahlenwert. in die wurzel nun losgelöst und im einsatz – magnetischen primes sind dabei solch galileo.
SLIDE 36
TRAINING DATA
supervised unsupervised
SLIDE 37
DEEP LEARNING
CREATE IMAGES LEARN FEATURES LEARN CLASSIFIERS PROCESS ALMOST EVERYTHING
SLIDE 38
DEEP LEARNING
NEED DATA (A LOT) A BLACK BOX EASY TO LEARN HARD TO MASTER PROCESS ALMOST EVERYTHING
SLIDE 39
DEEP LEARNING
TEUBI @DasTeutelbier teubi@chaos.social NUSSSCHALE PODCAST science in a german nutshell nussschale-podcast.de
listening to
N U S S S C H A L E P O D C A S T
SLIDE 40
DEEP LEARNING
TEUBI @DasTeutelbier teubi@chaos.social NUSSSCHALE PODCAST science in a german nutshell nussschale-podcast.de
listening to
N U S S S C H A L E P O D C A S T