[SYMBOL][A][B][C][D][E][F][G][H][I][J][K][L][M][N][O][P][R][S][T][U][V][W][Y][Z]
absolute value
academic papers
accuracy
activation functions
adding to layer
computable
continuous and infinite in domain
defined
hidden-layer
installation instructions
monotonic
nonlinear
output layer
similar inputs
slope
softmax computation
upgrading MNIST network
activation input parameter
actual error
__add__ function
addition backpropagation
additional functions, adding support for
algorithms
alpha
Anaconda framework
AND operator
arbitrary length
backpropagation with
challenge of
forward propagation with
weight update with
architecture of neural networks
importance of visualization tools
language
overview of
artificial neural networks
attenuation
autograd (automatic gradient computation)
adding cross entropy to
adding indexing to
general discussion
implementing LSTM with
to train neural network
upgrading to support multiuse tensors
used multiple times
automatic optimization.
See also autograd (automatic gradient computation).
adding
adding support
for additional functions
for negation
addition backpropagation
deep learning framework, 2nd
dynamic computation graph
layers
adding support for layer types
containing layers
cross-entropy layer
embedding, 2nd
in Keras or PyTorch.
loss-function layers
nonlinearity layers
recurrent neural network (RNN) layer
tensors
defined
that are used multiple times
averaged word vectors, RNN
Babi dataset
backpropagation
addition
in code
iteration of
recurrent neural network (RNN), 2nd
truncated
disadvantages of
iterating with
weighted average delta
weighted average error
.backward( ) method, 2nd, 3rd, 4th, 5th
bash commands
batch gradient descent, 2nd
batch_loss
batch_size, 2nd
batch_size/alpha pair
Bernoulli distribution
blogging, to teach deep learning
Bongo Java
bptt variable, 2nd, 3rd
calculus
cell-update vector
character language modeling.
See also LSTM (long short-term memory).
ciphertext
cluster labels
comparing
mean squared error, 2nd
measuring error
computable activation functions
computation graphs, 2nd
concatenation
conditional (sometimes) correlation, 2nd
continuous functions
convolutional neural networks
convolutional layer
implementation in NumPy
reusing weights in multiple places
corners.
See convolutional neural networks.
correlation
coefficients
creating
indirect
input/output
learning
negative
searching for
selective
correlation summarization
counting-based learning.
See nonparametric learning.
creation_op add
creators attribute
cross communication, 2nd
cross-entropy class, 2nd, 3rd, 4th
cross-entropy layer
cryptography
curves
data, grouping
data patterns
datapoints, 2nd
datasets
Babi
clustering into groups
IMDB movie reviews
learning whole
MNIST, 2nd, 3rd
preparing data
streetlight problem
transforming
debugging frameworks
decoder weight matrix
deep learning
adapted for beginning learners
analogies and
defined
difficulty level for learning
ongoing education
overview
project-based teaching method
reasons for
incremental automation of intelligence
potential for automation of skilled labor
stimulates intelligence and creativity
requirements for
high school–level mathematics
Jupyter Notebook
NumPy Python library
personal challenge to solve
Python knowledge
subfield of machine learning
teaching
textbook for
to understand frameworks and code libraries
deep learning framework, 2nd.
See also automatic optimization.
deep neural network
backpropagation
in code
iteration of
weighted average delta
weighted average error
batch gradient descent
building
correlation
creating
indirect
learning
full gradient descent
importance of
learning whole dataset
linear versus nonlinear networks
making predictions
matrices and matrix relationship
importing matrices into Python
patterns
streetlight problem
overfitting
preparing data
running program
sometimes correlation, 2nd
stacking
stochastic gradient descent, 2nd
weight pressure
conflicting pressure
weight update
defeat_delta variable
delta, multiplying by slope
delta variable, 2nd, 3rd, 4th, 5th
deniability, 2nd
derivatives
calculating
defined
example
relationship between weight and error and, 2nd
using to learn
weight_delta
diagonal
diff variable
direct imitation
direction_and_amount variable
discorrelation
divergence, 2nd
dot products, 2nd, 3rd, 4th
neural prediction
visualizing
double negative
down pressure
dropout technique, regulation
in code
evaluated on MNIST
general discussion
dynamic computation graph
DyNet framework
early stopping, 2nd
edges.
See convolutional neural networks.
ele_mul function, 2nd
elementwise addition, 2nd
elementwise multiplication, 2nd
elementwise operation
embedding layers, 2nd, 3rd
encryption
error curve
errors
error attribution
mean squared error, 2nd
measuring
positive
reducing
Euclidian distance
execution and output analysis, RNN
expand function, adding support for
exploding gradient problem
countering with LSTM
general discussion
toy example
federated learning
general discussion
hacking into
homomorphically encrypted
overview of, 2nd
fill-in-the-blank task, neural networks, 2nd
for loop, 2nd, 3rd, 4th
forward( ) method, 2nd
forward propagation
finishing with .index_select( ) method.
importance of visualization tools
linking variables
with multiple inputs
how it works
runnable code
weighted sum
with multiple outputs, 2nd
predicting with multiple inputs
using single input
neural networks
defined
purpose of
simple
stacking
NumPy Python library
overview of, 2nd, 3rd, 4th, 5th
predicting on predictions
prediction
recurrent neural network, 2nd
side-by-side visualization
frameworks, debugging
full gradient descent
functions
gates
generalization, regulation
get_parameters( ) method
global correlation summarization
goal_pred variable
goal_prediction variable
GPUs
gradient descent.
See also neural learning.
breaking
general discussion
iteration of
with multiple inputs
freezing one weight, 2nd
single-weight neural network versus
turning single delta into three weight_delta values
with multiple inputs and outputs
gradient descent generalizes to arbitrarily large networks
neural networks making multiple predictions using single input
with multiple outputs
visualizing dot products (weighted sums)
weights
graphs, 2nd
grouping data, 2nd, 3rd
Gryffindor
hidden variable
hidden-layer activation functions
hidden-to-hidden layer
hidden-to-output layer
high-low pattern
homomorphic encryption, 2nd
hot and cold learning
characteristics of
example
general discussion
ICML (International Conference on Machine Learning)
identity matrices, 2nd, 3rd
identity vectors
image classification
images matrix
IMDB movie reviews dataset
imitation
inceptionism
.index_select( ) method
indices array, 2nd
indirect correlation
indirect imitation
infinite functions
infinite parameters
input -> goal_prediction pairs, 2nd
input data pattern
input datapoints
input datasets, 2nd
input layers
input node
input values, 2nd
input variable, 2nd
input vector, 2nd, 3rd
input/output correlation
inputs
gradient descent with
freezing one weight
general discussion
generalizes to arbitrarily large networks
neural networks making multiple predictions using only single input
single-weight neural network vs.
turning single delta (on node) into three weight_delta values
how it works
overview of
runnable code
weighted sums
input-to-hidden layer
installation instructions, activation functions
intelligence targeting
intermediate dataset
intermediate predictions
International Conference on Machine Learning (ICML)
Keras framework, 2nd, 3rd
kernel_output
kernels
knob_weight variable, 2nd
labels, 2nd, 3rd, 4th, 5th
language, neural networks understanding
embedding layer
fill-in-the-blank task
general discussion
IMDB movie reviews dataset
interpreting output
loss function
meaning of neuron
natural language processing
neural architecture
predicting movie reviews
supervised natural language processing
word analogies
word correlation
word embeddings
Lasagne framework
Layer class
layer_0_delta variable
layer_1_delta variable, 2nd
layer_2_delta variable
layers
adding activation functions to
adding support for layer types
containing layers
cross-entropy layer
dimensionality of matrices and
embedding, 2nd
embedding layer translates indices into activations
in Keras or PyTorch
loss-function layers
nonlinearity layers
linear neural networks
linear nodes
list objects
local correlation summarization
log function
logical analysis
logical AND
long short-term memory.
See LSTM (long short-term memory).
loss function, 2nd, 3rd, 4th
loss.backward( ) function
loss-function layers
lossless representation
lower weights
LSTM (long short-term memory)
character language model
training
tuning
upgrading
countering vanishing and exploding gradients with
gates
using autograd system to implement
machine learning
make_sent_vect function
matrices and matrix relationship
importing matrices into Python
layers and
patterns
streetlight problem
matrix multiplication, adding support for
max pooling
mean pooling
mean squared error, 2nd, 3rd, 4th, 5th, 6th, 7th
measuring error
memorization, regulation
memorizing neural network code
mini-layers
missing values
MNIST (Modified National Institute of Standards and Technology) dataset
overview of
three-layer network on
upgrading
MNIST digit classifier
MNISTPreprocessor notebook
monotonic activation functions
multi-input gradient descent
multiple inputs
gradient descent with
freezing one weight
general discussion
generalizes to arbitrarily large networks
neural networks making multiple predictions using only single input
single-weight neural network versus
turning single delta (on node) into three weight_delta values
how it works
runnable code
weighted sums
multiple outputs
gradient descent with
generalizes to arbitrarily large networks
neural networks making multiple predictions using only single input
how it works
predicting with multiple inputs
using single input
multiplication function, adding support for
n linear layers
n output neurons
n_batches
n_hidden parameter
n_layers input parameter
Nanodegree
NaNs (not-a-numbers)
natural language processing.
See NLP.
n-dimensional tensors
__neg__ function
negation, adding support for
negative correlation
negative derivatives
negative labels, 2nd, 3rd
negative numbers
negative reversal attribute, 2nd, 3rd
negative sampling
negative sensitivity
negative weight
neural architecture.
See architecture of neural networks.
neural learning
alpha
calculus and
comparing
mean squared error, 2nd
measuring error
derivatives
calculating
defined
example
relationship between weight and error and, 2nd
using to learn
weight_delta
divergence
error attribution
functions
gradient descent
breaking
general discussion
iteration of
hot and cold learning
characteristics of, 2nd, 2nd
memorizing
overcorrections, 2nd
visualizing
reducing error
steps of
neural networks.
See also deep neural network; ; neural learning.
backpropagation
in code
iteration of
weighted average delta
weighted average error
batch gradient descent
building
correlation
creating
indirect
learning
defined
full gradient descent
importance of
learning whole dataset
linear versus nonlinear networks
making multiple predictions using single input
making predictions
matrices and matrix relationship
importing matrices into Python
patterns
streetlight problem
overfitting
preparing data
purpose of
running program
simple
sometimes correlation, 2nd
stacking, 2nd
stochastic gradient descent, 2nd
visualizing
architecture
correlation summarization
importance of visualization tools
side by side
simplifying, 2nd
vector-matrix multiplication
weight pressure
conflicting pressure
weight update
neural prediction
with multiple inputs
how it works
runnable code
weighted sum
with multiple outputs, 2nd
predicting with multiple inputs
using single input
neural networks
defined
purpose of
simple
stacking
NumPy Python library
predicting on predictions
prediction
neurons, 2nd, 3rd
NLP (natural language processing), 2nd, 3rd
nodes
noise, 2nd, 3rd.
See also regulation.
nonlinear activation functions
nonlinear neural networks
nonlinearities.
See also activation functions.
nonlinearity layers, 2nd
nonparametric learning
counting-based methods
parametric learning versus
normalization
normalized variants
normed_weights matrix
NOT operator
not-a-numbers (NaNs)
np.dot function
NumPy Python library, 2nd, 3rd
objective function
one_hot utility matrix
one-dimensional tensors
one-hot encoding
on-the-job training
open source project
OpenMined, 2nd
OR operator
output data pattern
output datapoints
output datasets, 2nd
output layer activation functions
choosing
configurations
no activation function
sigmoid
softmax
outputs
converting to slope
gradient descent with
generalizes to arbitrarily large networks
neural networks making multiple predictions using only single input
how it works
neural networks
predicting with multiple inputs
using single input
overcorrections
alpha
visualizing
overfitting
causes of
general discussion
overview of
overshooting
parameters, 2nd, 3rd
parametric learning
nonparametric learning versus
supervised
unsupervised
patterns
peer support, for deep learning
perplexity metric
pip install phe command
pixels, 2nd, 3rd, 4th
plaintext
plausible deniability, 2nd
pooling
positive errors
positive labels, 2nd, 3rd
positive sensitivity
practice, importance of
pred variable
predictions
deep neural networks
with multiple inputs
how it works
runnable code
weighted sum
with multiple outputs, 2nd
predicting with multiple inputs
using single input
neural networks
defined
purpose of
simple
stacking
NumPy Python library
predicting images
predicting movie reviews
predicting on predictions
prediction
privacy.
See also security and privacy.
federated learning
general discussion
hacking into
homomorphically encrypted
homomorphic encryption
privacy
secure aggregation
spam
private key
probabilities.
See activation functions.
project-based teaching method
propagation
finishing with .index_select( ) method
importance of visualization tools
linking variables
with multiple inputs
how it works
runnable code
weighted sum
with multiple outputs, 2nd
predicting with multiple inputs
using single input
neural networks
defined
purpose of
simple
stacking
NumPy Python library
overview of, 2nd, 3rd, 4th, 5th, 6th
predicting on predictions
prediction
recurrent neural network, 2nd
side-by-side visualization
public key
pure error, 2nd
Python
creating sentence embeddings using identity matrices in
forward propagation in
learning
NumPy Python library, 2nd, 3rd
Python Codecademy course
PyTorch, 2nd, 3rd
random subsections
randomized response
randomness
raw error, 2nd
recurrent embeddings
recurrent matrix
recurrent neural network.
See RNN.
reducing error
regularization, 2nd, 3rd, 4th
regulation
batch gradient descent
dropout technique
in code
evaluated on MNIST
general discussion
early stopping
generalization
memorization
overfitting
causes of, 2nd
predicting images
three-layer network on MNIST
relu function, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th, 9th
relu2deriv function, 2nd, 3rd
reviews2vectors matrix
RNN (recurrent neural network)
arbitrary length of data
backpropagation with
challenge of
forward propagation with
weight update with
averaged word vectors
Babi dataset
backpropagation
character language modeling
comparing sentence vectors
execution and output analysis
forward propagation in Python
overview of, 2nd
sentence embeddings
setting up
vanishing and exploding gradients
word embeddings
how information is stored in
limitations of
neural networks use of
summing
RNNCell class
runnable code, neural prediction
running.backward( ) method
sampling output
scalar multiples
scalar-matrix multiplication
scalars
scaling attribute, 2nd, 3rd
security and privacy
federated learning
general discussion
hacking into
homomorphically encrypted
homomorphic encryption
privacy
secure aggregation
spam
selective correlation
self.children counter
self.data array
self.w_hh layer
self.w_ho layer
self.w_ho.forward(h)
self.w_ih layer
self.weight
sensitivity, 2nd, 3rd
sent2output layer
sentence embeddings, RNN
sentence vectors, 2nd
transition matrices
sentiment dataset
Sequential( ) method
SGD class
shape, 2nd, 3rd
sharpness of attenuation
short-term memory
sigmoid( ) function, 2nd, 3rd, 4th, 5th, 6th, 7th
signal, 2nd.
See also regulation.
similarity
simple neural networks
simplifying visualization, 2nd
single-input gradient descent
single-weight neural network
slope
converting output to
multiplying delta by
overview of
softmax computation
activation functions
output layer activation functions
softmax function, 2nd, 3rd, 4th, 5th
sometimes (conditional) correlation, 2nd
sox_delta variable
spam
square matrix
squiggly line
stacked convolutional layers
stacking neural networks, 2nd
stacks of layers
static computation graph
step_amount variable
stickiness
stochastic gradient descent optimizer, 2nd, 3rd, 4th
stopping attribute, 2nd, 3rd
streetlight problem
datasets
matrices and matrix relationship
subtraction function, adding support for
sum function, adding support for
sum pooling
.sum(dim) function
summing embeddings
supervised machine learning
supervised natural language processing (NLP)
supervised parametric learning
tanh( ) function, 2nd, 3rd, 4th, 5th
target labels, meaning of neuron based on
tasks, NLP
Tensor class, 2nd
Tensor.backward( ) function
TensorFlow, 2nd
tensors
adding nonlinear functions to
automatic gradient computation (autograd)
defined
that are used multiple times
test( ) function
test_images variable
test_labels variable
testing accuracy
Theano
three-layer networks, 2nd
topic classification
train( ) function
training accuracy
Training-Acc
transition weights
transpose function, adding support for
trial and error, 2nd.
See also parametric learning.
true signal
truncated backpropagation
disadvantages of
iterating with
overview of
Twitter
two-dimensional tensors
two-layer networks, 2nd
<unk> tokens
unsupervised machine learning, 2nd
up pressure
utility functions
validation set
vanishing gradient problem
countering with LSTM
general discussion
toy example
variable-length text
variables
linking
multiplying
variants
vect_mat_mul function
vectors, 2nd, 3rd, 4th, 5th
vector-scalar addition and multiplication
virtual graph
visualizing neural networks
architecture
correlation summarization
importance of visualization tools
side by side
simplifying, 2nd
vector-matrix multiplication
defined
letters can be combined to indicate functions and operations
linking variables
using letters instead of pictures
volume, 2nd
w_sum function, 2nd
weight pressure
conflicting pressure
weight update
Weight Pressure table, 2nd
weight values, 2nd
weight variable
weight vector
weight_delta variable, 2nd, 3rd, 4th, 5th
weighted average delta
weighted average error
weighted sums, 2nd
neural prediction
visualizing
weights.
See also multiple inputs.
batch gradient descent
convolutional neural networks
freezing one weight
full gradient descent
MNIST dataset
stochastic gradient descent
turning single delta into three weight_delta values
visualizing weight values
weight update with arbitrary length
weights variable, 2nd, 3rd, 4th, 5th
weights vector
wlrec predictor
word analogies
word correlation, capturing in input data
word embeddings
comparing
recurrent neural network
how information is stored in
limitations of
neural networks use of
summing
word vectors, 2nd
3.16.139.8