Conscious Artificial Intelligence Activity
It's possible for machines to become self-aware.
Status: Beta
Brought to you by:
jpss
Activity for Conscious Artificial Intelligence
-
Joao Paulo Schwarz Schuler committed [r1994] on Current Source Code
experiment for numerical stability
-
add FilterCSVWithNumbersUpToMax
-
experimental transformer code
-
fixes ReverseGroupedOneHotEncoding
-
fixes TNNetNeuron.InitAdam
-
fixes GroupedOneHotEncoding
-
fixes access violation error
-
fixes access violation error
-
NLP coding
-
New experimental CAI self attention
-
MulWeights is now a function
-
fixes TNNetSigmoid.Backpropagate()
-
InitHeUniformForAllDenseLayers
-
Adding InitGlorotBengioUniformForAllConvLayers
-
CAI is again using He for convolutional layers
-
Convolutional layers got the same initialization as Keras
-
Updating CAI Transformer
-
Updating CAI Transformer
-
Updating CAI Transformer
-
fixes pointwise softmax with no forward and skip derivative
-
adding debug code
-
adding debug code
-
adding debug code
-
better normalization methods
-
fixes backpropagation with branching
-
fixes backpropagation with branching
-
adds plenty of self testing
-
fixes backpropagation with branching
-
New TNNetMovingStdNormalization learning algorithm
-
New TNNetMovingStdNormalization learning algorithm
-
better delta clip
-
inertia coding - better debug
-
inertia coding - better debug
-
Clear delta and inertia at loading
-
More debug info
-
Save best accuracy x loss
-
Save best accuracy x loss
-
Better preperties usage
-
adding validation error and loss properties to NFIT
-
better initialization parameters for NLP
-
updating default optimizer values
-
Adding local connect linear
-
fixes pixel class reading
-
new loss functions
-
better log messages
-
fixing one-hot encoding
-
small optimization for noforward in dotproducts
-
fixes noforward parameter in dotproducts
-
optimizing dotproducts backprop
-
coding grouped pointwise softmax
-
Grouped one-hot encoding
-
testing transformers
-
coding NLP support
-
coding NLP support
-
No forward calculation at softmax
-
adding debug functions
-
More safety checks
-
Optimizing TVolume.CopyTransposingAs2D
-
Fixing departing branches count
-
Fixing departing branches count
-
Fixing DotProducts departing branches
-
Fixing DotProducts departing branches
-
closer implementation of transformer decoder
-
closer implementation of transformer encoder
-
fixes embedding learning direction
-
fixes embedding learning direction
-
removing min learning rate
-
removing extra transpose from transformers
-
experimental max norm in CAI transformer
-
In testing we trust - fixing initialization of embedding
-
In testing, we trust: fixes PointwiseSoftMax
-
In testing, we trust: fixes PointwiseSoftMax
-
protects against overflow with adam
-
protects against overflow with adam
-
in code review we trust
-
in code review we trust
-
updating SGD optimizer
-
embedding now uses uniform initialization
-
fixing broken softmax
-
in testing/replication we trust
-
coding adam
-
coding adam
-
coding adam
-
fixing introduced bug and speeing up
-
experimenting delta norm
-
experimenting delta norm
-
in testing we trust
-
in testing we trust
-
OneHotEncodingReversed
-
testing/bug fixing transformers
-
testing/bug fixing transformers
-
Coding transformers - missed debug
-
Coding transformers
-
coding embedding
-
coding token embedding
-
Adding Transformer Example
-
coding TNNetTokenAndPositionalEmbedding
-
coding TNNetTokenAndPositionalEmbedding
-
coding token embedding
-
updating source code comments