MyCaffe 0.11.3.25-beta1

MyCaffe AI Platform and Test Application (CUDA 11.3, cuDNN 8.2) with Seq2Seq and Attention Support.

CUDA 11.3, cuDNN 8.2, nvapi 465, Windows 20H2, Driver 466.27

MyCaffe[1] (a complete C# re-write of CAFFE[2]) now supports Seq2Seq with Attention!

IMPORTANT NOTE: When using TCC mode, we recommend that ALL headless GPU’s are placed in TCC mode for we have experienced stability issues when using a mix of TCC and WDM modes with headless GPU’s.

REQUIRED SOFTWARE to use MyCaffe: 1.) Download and install full version of Microsoft SQL Express 2016 (or later). NOTE: The full version of SQL Express must installed as opposed to the light version included in Visual Studio. Microsoft SQL Express can be downloaded from https://www.microsoft.com/en-us/sql-server/sql-server-downloads

REQUIRED SOFTWARE to build MyCaffe: 1.) Install NVIDIA CUDA 11.3 which you can download from https://developer.nvidia.com/cuda-downloads 2.) Install NVIDIA cuDNN 8.2 which you can download from https://developer.nvidia.com/cudnn

This release of the MyCaffe AI Platform and Test Applications has the following new additions: • CUDA 11.3/cuDNN 8.2 supported (with driver 466.27 or above). • Windows 20H2, OS Build 19042.985 now supported. • Added ability to TestMany after specified time. • Added new ModelGym support for dual models with RNN • Added mid-point analysis to TestMany. • Added new MISH Activation Layer • Added new MAE Loss Layer • Added new Merge Layer for use with LSTM models. • Added new HDF5Data Layer. • Added new Attention Layer • Added new LSTMAttention Layer for Seq2Seq models. • Added new Copy Layer for use with LSTM models. • Added new Seq2Seq sample to MyCaffe-Samples

The following bug fixes are in this release: • Fixed bug in concat_dim, now set to uint? and ignored when null. • Fixed bugs in Rnn layers related to Workspace and Reserved, changed to ulong. • Fixed bugs related to running special tests with older CUDA versions. • Fixed bugs related to Reshape in LSTM_SIMPLE layer. • Fixed bugs in SimpleDatum AccumulateMean when used on byte data. • Fixed bugs related to accuracy calculation when learning RNN's. • Fixed bug related to loading debug and criteria data from Virtual image. • Fixed LaTex errors in online help. • Fixed bug in SoftmaxLoss layer where axis not properly transferred to softmax. • Fixed bug where InnerProduct transpose parameter was not parsed correctly. • Improved image queries times from physical database. • Improved loading raw image times.

Easily run Seq2Seq[3] models with Attention[4], Single-Shot Multi-Box Nets[5][6], import/export ONNX AI Models, run Triplet Nets[7][8], run Siamese Nets[10][11], Neural Style, train Deep Q-Learning or Policy Gradient models to beat Pong or Cart-Pole, or create the CIFAR-10 and MNIST datasets using the MyCaffe-Samples (https://github.com/MyCaffe/MyCaffe-Samples) and MyCaffe Test Application which you can download from the MyCaffe GitHub site.

Schedule distributed AI work packages, or create and train Single-Shot Multi-Box[5][6], Triplet Net[7][8], Siamese Net[9][10], Deep Q-Learning with NoisyNet and Experienced Replay, Policy Gradient, Neural Style Transfer, Recurrent Learning, Policy Gradient Reinforcement Learning, Auto-Encoder, DANN and ResNet models by following step-by-step instructions in the SignalPop Tutorials. And, to see other cool examples that show what MyCaffe can do, see the SignalPop Examples.

If you would like to visually design, develop, test and debug your models, see the SignalPop AI Designer specifically designed to enhance your MyCaffe deep learning.

Also, check out the SignalPop Universal Miner that not only keeps your GPU's cool as you train, but also gives you detailed information on each of your GPU's (such as temperature, fan speed, overclock, and usage), and allows you to easily mine Ethereum. When not training AI, put those GPU's to use making some Ether - never let a good GPU go to waste!

Happy ‘deep’ learning!

[1] MyCaffe: A Complete C# Re-Write of Caffe with Reinforcement Learning by D. Brown, 2018.

[2] Caffe: Convolutional Architecture for Fast Feature Embedding by Yangqing Jia, Evan Shelhamer, Jeff Donahue, Sergey Karayev, Jonathan Long, Ross Girshick, Sergio Guadarrama, Trevor Darrell, 2014, arXiv:1408.5093

[3] Attention - Seq2Seq Models by Pranay Dugar, Toward Data Science, 2019.

[4] Attention Is All You Need by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin, 2017, ArXiv:1706.03762

[5] SSD: Single Shot MultiBox Detector by Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg, 2016.

[6] GitHub: SSD: Single Shot MultiBox Detector, by weiliu89/caffe, 2016

[7] Deep metric learning using Triplet network by Elad Hoffer and Nir Ailon, 2018, arXiv:1412.6622

[8] In Defense of the Triplet Loss for Person Re-Identification by Alexander Hermans, Lucas Beyer and Bastian Liebe, 2017, arXiv:1703.07737v2

[9] Siamese Network Training with Caffe by Berkeley Artificial Intelligence (BAIR)

[10] Siamese Neural Network for One-shot Image Recognition by G. Koch, R. Zemel and R. Salakhutdinov, ICML 2015 Deep Learning Workshop, 2015.