federica bianco
astro | data science | data for good
what's behind facial recognition: technology and bias
University of Delaware
Department of Physics and Astronomy
federica bianco
Biden School of Public Policy and Administration
Data Science Institute
who needs to learn
Educate Policy makers
without understanding how ML works policy makers do not have the instruments to regulate it
Education for the people
but does this put the burden on the victims?
Educating DS practitioners in communicating DS concepts
the put the burden back on the practitioners
Datascience Education to Help and Protect us
Jack Dorsey (Twitter CEO) at TED 2019
boring the TED audience with details
Zuckerberg (Facebook CEO) deflecting questions at senate hearing
#UDCSS2020
@fedhere
Data Science is a black box
Models are neutral, data is biased
two dangerous data-ethics myths
#UDCSS2020
@fedhere
Data Science is a black box
machine learning models are
Epistemic transparency
tration by Hanne Morstad
Accountability: who is responsible if an algorithm does harm
[Machine Learning is the] field of study that gives computers the ability to learn without being explicitly programmed.
Arthur Samuel, 1959
what is a ML?
a model is a low dimensional representation of a higher dimensionality datase
what is a "model" in ML?
what is machine learning?
[Machine Learning is the] field of study that gives computers the ability to learn without being explicitly programmed.
Arthur Samuel, 1959
model parameters: slope, intercept
data
mathematical formula
what is machine learning?
ML: study, development, and applicaton of any model with parameters learnt from the data
Objective Function
time
time
time
which is the "best fit" line? A , B, C, D?
A
B
C
D
to select the best fit parameters we define a function of the parameters to minimize or maximize
Objective Function
L_1 = \sum_{i=1}^N|f(x) - y|
L_2 = \sum_{i=1}^N(f(x) - y)^2
x1
x2
to select the best fit parameters we define a function of the parameters to minimize or maximize
Objective Function
L_1 = \sum_{i=1}^N|f(x) - y|
L_2 = \sum_{i=1}^N(f(x) - y)^2
x1
x4
to select the best fit parameters we define a function of the parameters to minimize or maximize
Machine Learning models are parametrized representation of "reality" where the parameters are learned from finite sets of realizations of that reality
(note: learning by instance, e.g. nearest neighbours, may not comply to this definition)
Machine Learning is the disciplines that conceptualizes, studies, and applies those models.
Key Concept
what is machine learning?
used to:
- classify based on examples
- understand structure of feature space
- regression (classification with infinitely small classes)
- understand which features are important in prediction (to get close to causality)
General ML usage
Accountability
Accountability
Explainability
Accountability
Explainability
Fairness
Fairness
Accountability
Explainability
Fairness
Fairness
Privacy and data rights
What drives inference
Inference has different drievers in different times depending on the resources available and dominant culture
Accountability will look different in these different
Galileo Galilei 1610
Following: Djorgovski
https://events.asiaa.sinica.edu.tw/school/20170904/talk/djorgovski1.pdf
Experiment driven
what drives
inference
@fedhere
Enistein 1916
what drives
inference
Theory driven | Falsifiability
Experiment driven
@fedhere
Ulam 1947
Theory driven | Falsifiability
Experiment driven
Simulations | Probabilistic inference | Computation
http://www-star.st-and.ac.uk/~kw25/teaching/mcrt/MC_history_3.pdf
@fedhere
what drives
inference
what drives
astronomy
the 2000s
Theory driven | Falsifiability
Experiment driven
Simulations | Probabilistic inference | Computation
Big Data + Computation | pattern discovery | predict by association
@fedhere
data driven: lots of data, drop theory and use associations
algorithmic transparency
strictly policy issues:
proprietary algorithms + audability
#UDCSS2020
@fedhere
technical + policy issues:
data access and redress + data provenance
algorithmic transparency
https://www.darpa.mil/attachments/XAIProgramUpdate.pdf
trivially intuitive
generalized additive models
decision trees
SVM
Random Forest
Deep Learning
Accuracy
univaraite
linear
regression
algorithmic transparency
#UDCSS2020
@fedhere
we're still trying to figure it out
algorithmic transparency
https://www.darpa.mil/attachments/XAIProgramUpdate.pdf
trivially intuitive
generalized additive models
decision trees
SVM
Random Forest
Deep Learning
Accuracy in solving complex problems
univaraite
linear
regression
algorithmic transparency
#UDCSS2020
@fedhere
we're still trying to figure it out
algorithmic transparency
trivially intuitive
generalized additive models
decision trees
Deep Learning
number of features that can be effectively included in the model
thousands
1
SVM
Random Forest
univaraite
linear
regression
https://www.darpa.mil/attachments/XAIProgramUpdate.pdf
algorithmic transparency
#UDCSS2020
@fedhere
Accuracy in solving complex problems
we're still trying to figure it out
algorithmic transparency
trivially intuitive
univaraite
linear
regression
generalized additive models
decision trees
Deep Learning
SVM
Random Forest
https://www.darpa.mil/attachments/XAIProgramUpdate.pdf
time
algorithmic transparency
#UDCSS2020
@fedhere
Accuracy in solving complex problems
we're still trying to figure it out
algorithmic transparency
1
Machine learning: any method that learns parameters from the data
2
The transparency of an algorithm is proportional to its complexity and the complexity of the data space
3
The transparency of an algorithm is limited by our own ability and preparedness to interpret it
Toward Interpretable Machine Learning, Samek+2003
algorithmic transparency
#UDCSS2020
@fedhere
linear regression
algorithmic transparency
A single tree model
algorithmic transparency
1943
M-P Neuron McCulloch & Pitts 1943
M-P Neuron
Perceptrons are linear classifiers: makes its predictions based on a linear predictor function
combining a set of weights (=parameters) with the feature vector.
The perceptron algorithm : 1958, Frank Rosenblatt
x
y
1958
y ~= ~\sum_i w_ix_i ~+~ b
The perceptron algorithm : 1958, Frank Rosenblatt
+b
f
w_2
w_1
w_N
output
f
activation function
weights
w_i
bias
b
sigmoid
f
\sigma = \frac{1}{1 + e^{-z}}
.
.
.
x_1
x_2
x_N
y ~= f(~\sum_i w_ix_i ~+~ b)
Perceptrons are linear classifiers: makes its predictions based on a linear predictor function
combining a set of weights (=parameters) with the feature vector.
The perceptron algorithm : 1958, Frank Rosenblatt
Perceptron
The Navy revealed the embryo of an electronic computer today that it expects will be able to walk, talk, see, write, reproduce itself and be conscious of its existence.
The embryo - the Weather Buerau's $2,000,000 "704" computer - learned to differentiate between left and right after 50 attempts in the Navy demonstration
NEW NAVY DEVICE LEARNS BY DOING; Psychologist Shows Embryo of Computer Designed to Read and Grow Wiser
July 8, 1958
\vec{y} = f_N(....(f_1(\vec{x}{ W_i + b_1}...W_N + b_N)))
x1
x2
b1
b2
b3
b
w11
w12
w13
w21
w22
w23
multilayer perceptron
w: weight
sets the sensitivity of a neuron
b: bias:
up-down weights a neuron
EXERCISE
output
how many parameters?
input layer
hidden layer
output layer
hidden layer
facial recognition
how it works
#UDCSS2020
@fedhere
facial recognition
typically 68 landmarks
how it works
facial recognition
typically 68 landmarks
how it works
facial recognition
what its not:
Autoencoders
What do NN do? approximate complex functions with series of linear functions
.... so if my layers are smaller what I have is a compact representation of the data
Autoencoder Architecture
Feed Forward DNN:
the size of the input is 5,
the size of the last layer is 2
Unsupervised learning with
Neural Networks
What do NN do? approximate complex functions with series of linear functions
To do that they extract information from the data
Each layer of the DNN produces a representation of the data a "latent representation" .
The dimensionality of that latent representation is determined by the size of the layer (and its connectivity, but we will ignore this bit for now)
.... so if my layers are smaller what I have is a compact representation of the data
- Encoder: outputs a lower dimensional representation z of the data x (similar to PCA, tSNE...)
- Decoder: Learns how to reconstruct x given z: learns p(x|z)
Building a DNN
with keras and tensorflow
Trivial to build, but the devil is in the details!
Building a DNN
with keras and tensorflow
Trivial to build, but the devil is in the details!
from keras.models import Sequential
#can upload pretrained models from keras.models
from keras.layers import Dense, Conv2D, MaxPooling2D
#create model
model = Sequential()
#create the model architecture by adding model layers
model.add(Dense(10, activation='relu', input_shape=(n_cols,)))
model.add(Dense(10, activation='relu'))
model.add(Dense(1))
#need to choose the loss function, metric, optimization scheme
model.compile(optimizer='adam', loss='mean_squared_error')
#need to learn what to look for - always plot the loss function!
model.fit(x_train, y_train, validation_data=(x_test, y_test),
epochs=20, batch_size=100, verbose=1)
#note that the model allows to give a validation test,
#this is for a 3fold cross valiation: train-validate-test
#predict
test_y_predictions = model.predict(validate_X)Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
encoder
This autoencoder model has a 64-neuron bottle neck. This means it will generate a compressed representation of the data out of that layer which is 16-dimensional (the original size is 784 pixels)
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
This autoencoder model has a 64-neuron bottle neck. This means it will generate a compressed representation of the data out of that layer which is 16-dimensional (the original size is 784 pixels)
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
decoder
This autoencoder model has a 64-neuron bottle neck. This means it will generate a compressed representation of the data out of that layer which is 16-dimensional (the original size is 784 pixels)
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
This autoencoder model has a 64-neuron bottle neck. This means it will generate a compressed representation of the data out of that layer which is 16-dimensional (the original size is 784 pixels)
bottle neck
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
This simple odel has 200000 parameters!
My original choice is to train it with "adadelta" with a mean squared loss function, all activation functions are relu, appropriate for a linear regression
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
What should I choose for the loss function and how does that relate to the activation functiom and optimization?
Building a DNN
with keras and tensorflow
autoencoder for image recontstruction
What should I choose for the loss function and how does that relate to the activation functiom and optimization?
| loss | good for | activation last layer | size last layer |
|---|---|---|---|
| mean_squared_error | regression | linear | one node |
| mean_absolute_error | regression | linear | one node |
| mean_squared_logarithmit_error | regression | linear | one node |
| binary_crossentropy | binary classification | sigmoid | one node |
| categorical_crossentropy | multiclass classification | sigmoid | N nodes |
| Kullback_Divergence | multiclass classification, probabilistic inerpretation | sigmoid | N nodes |
resources
Neural Network and Deep Learning
an excellent and free book on NN and DL
http://neuralnetworksanddeeplearning.com/index.html
Deep Learning An MIT Press book in preparation
Ian Goodfellow, Yoshua Bengio and Aaron Courville
https://www.deeplearningbook.org/lecture_slides.html
History of NN
https://cs.stanford.edu/people/eroberts/courses/soco/projects/neural-networks/History/history2.html
models are neutral, the bias is in the data (or is it?)
Why does this AI model whitens Obama face?
Simple answer: the data is biased. The algorithm is fed more images of white people
models are neutral, the bias is in the data (or is it?)
Why does this AI model whitens Obama face?
Simple answer: the data is biased. The algorithm is fed more images of white people
But really, would the opposite have been acceptable? The bias is in society
Joy Boulamwini
models are neutral, the bias is in the data (or is it?)
accountability
- can scientists be held responsible?
- should whoever commissions be responsible?
- is nobody responsible under the premise that decisions are objective? -> are they objective?, what does objective mean?, how can we objectively measure objectivity
accountability
because of its complexity FR is cmmissioned to specialists
accountability
In a press release, the ACLU wrote, “Mr. Williams’ experience was the first case of wrongful arrest due to facial recognition technology to come to light in the United States.”
accountability
In a press release, the ACLU wrote, “Mr. Williams’ experience was the first case of wrongful arrest due to facial recognition technology to come to light in the United States.”
Who is responsible for setting the threshold?
FR returns a probabilistic result
a threshold is chosen to turn it into a T/F match for decision making
Decide which model is appropriate (depends on data and question)
where is the bias?
#UDCSS2020
@fedhere
1 - model selection
we are still trying to figure it out
we are still trying to figure it out
trivially intuitive
generalized additive models
decision trees
SVM
Random Forest
Deep Learning
Accuracy
univaraite
linear
regression
where is the bias?
Decide what your target function is
Machine learning models are functions that "learn" their parameters from the data.
They "learn" by minimizing or maximize some quantity.
What should you minimize?
#UDCSS2020
@fedhere
https://towardsdatascience.com/machine-learning-fundamentals-via-linear-regression-41a5d11f5220
2 - cost function
where is the bias?
They "learn" by minimizing or maximize some quantity.
What should you minimize?
the hypothetical trolley problem suddenly is real
self-driving cars
#UDCSS2020
@fedhere
2 - cost function
where is the bias?
They "learn" by minimizing or maximize some quantity.
What should you minimize?
prosecutorial justice
minimize number of people incarcerated unjustly
maximize public safety
#UDCSS2020
@fedhere
OR
2 - cost function
Explore the data
discover some of the bias
(trust me, there is more!)
it's not easy
there's covariance
missing data
where is the bias?
#UDCSS2020
@fedhere
3 - data selection and preparation
remove the bias...
(few try)
#UDCSS2020
@fedhere
where is the bias?
3 - data selection and preparation
Machine learning learns from examples... what if the examples are ... racist?
GPT-3
unethical applications of FR
https://modelviewculture.com/pieces/the-hidden-dangers-of-ai-for-queer-and-trans-people
unethical applications of FR
unethical applications of FR
Text
unethical applications of FR
social protests
https://www.washingtonpost.com/technology/2020/06/12/facial-recognition-ban/
Policy and Resistance
https://www.washingtonpost.com/technology/2021/02/17/facial-recognition-biden/
https://www.washingtonpost.com/technology/2019/05/22/blasting-facial-recognition-technology-lawmakers-urge-regulation-before-it-gets-out-control/
key concepts
MACHINE LEARNING
- Machine Learning models are parametrized representation of "reality" where the parameters are learned from finite sets of realizations of that reality
- Unsupervised learning: all variables observed for all data, looking for natural grouping of datapoints in the N-dim space
- Supervised learning: a target variable is known for (a subset of) the data and the goal is to predict it for new (the rest of the) data
DATA ETHICS
- epistemic transparency:not all models are the same
- there is a tradeoff between epistemic transparency and the ability to handle complex data
- The bias enter data science in (at least) data; model selection; target function and optimization choices; validation
Text
thank you!
#UDCSS2020
@fedhere
University of Delaware
Department of Physics and Astronomy
federica bianco
Biden School of Public Policy and Administration
Data Science Institute
fbianco@udel.edu
https://www.tandfonline.com/doi/full/10.1080/1369118X.2018.1477967
https://medium.com/payoff/transparency-in-data-science-9a8778083b3
https://www.americanscientist.org/article/a-peek-at-proprietary-algorithms
https://medium.com/payoff/transparency-in-data-science-9a8778083b3
https://weaponsofmathdestructionbook.com/
The Ethics of the Ethics of AI https://www.oxfordhandbooks.com/view/10.1093/oxfordhb/9780190067397.001.0001/oxfordhb-9780190067397-e-2 , Thomas M. Powers and Jean-Gabriel Ganascia
ethicsds_bianco
By federica bianco
