SLIDE 1
Jenna Zeigen (she/her) QueensJS 8/5/2020
- rganizer of
EmpireJS
- rganizer of
how to train your model Jenna Zeigen (she/her) QueensJS 8/5/2020 - - PowerPoint PPT Presentation
how to train your model Jenna Zeigen (she/her) QueensJS 8/5/2020 - - PowerPoint PPT Presentation
how to train your model Jenna Zeigen (she/her) QueensJS 8/5/2020 senior frontend engineer at Slack organizer of BrooklynJS organizer of EmpireJS @zeigenvector jenna.is/at-queensjs-2020 machine learning creating algorithms that improve
SLIDE 2
SLIDE 3
jenna.is/at-queensjs-2020
@zeigenvector
SLIDE 4
creating algorithms that improve automatically through experience over time
machine learning
SLIDE 5
building a mathematical model based on "training data" in
- rder to make
predictions
machine learning
SLIDE 6
when the answers are known ahead of time, and the computer tries to find a model to fit the data
supervised learning
i.e. classification
SLIDE 7
when the answers aren't known ahead of time, and computer finds patterns
unsupervised learning
i.e. clustering
SLIDE 8
when the answers aren't known ahead of time and the algorithm learns by trial and error through "incentives"
reinforcement learning
popular in teaching computers to play games
SLIDE 9
it says "math"
SLIDE 10
it's the matrix because there are a lot of matrices in ml
SLIDE 11
put objects into groups based on their characteristics
classifiers
haha no math yet
SLIDE 12
Do this based on a boundary that is a "line" (or through ~*linear combination*~)
linear classifiers
- k now math
SLIDE 13
linear classifier
haha
SLIDE 14
autocomplete ranking is a matter of classification — is it the thing you're looking for
- r not?
how we train our model
SLIDE 15
tl;dr, turning logs into decimals using supervised learning
how we train our model
SLIDE 16
how we train our model
"features" are the attributes of the item that could be influencing the classification
feature extraction
SLIDE 17
how we train our model
creates a "feature vector" for each item, a list
- f all the
features and their values
feature extraction
p.s. even images can be represented as vectors
SLIDE 18
how we train our model
feature vectors from the selected and not selected items are used as data to train the "model"
training
SLIDE 19
how we train our model
the model will be a vector which has weights for each of the features
training
SLIDE 20
how we train our model
training
SLIDE 21
✅ ❌ ✅ ✅ ✅ ✅ ✅ ✅ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌ ❌
🔦
❌ ❌ ❌ ❌ ❌ ✅ ❌ ❌ ❌ ❌ ❌ ❌
how we train our model
training
(except this is a multidimensional space and this is a hyperplane not a line lol humans 🧡)
SLIDE 22
vonktor
how we train our model
training
minimize costs $$$!!!
SLIDE 23
how we train our model
An item's score is the sum of the product of each feature's value and its weight
scoring
For MA TH , head to https://en.wikipedia.org/wiki/Linear_classifier
SLIDE 24
https://towardsdatascience.com/understanding-neural-networks-19020b758230
"neural networks"
- k but what about
~*deep learning*~
🔦
hotness
SLIDE 25
https://towardsdatascience.com/understanding-neural-networks-19020b758230
- k but what about
~*deep learning*~
that's deep
SLIDE 26
https://towardsdatascience.com/understanding-neural-networks-19020b758230
- k but what about
~*deep learning*~
worth the weight
SLIDE 27
https://towardsdatascience.com/understanding-neural-networks-19020b758230
- k but what about
~*deep learning*~
minimize cost, like before
SLIDE 28
ethically.
how to train your model
🌷
spiciness?
SLIDE 29
algorithmic bias is real.
how to train your model
🌷
spiciness?
SLIDE 30
thanks!
jenna.is/at-queensjs-2020