DeepLearn.js 0.3.0 release, add support for iOS

DeepLearn.js 0.3.0 has been released, this is a hardware-accelerated machine learning JavaScript library that can run entirely in the browser without installation and back-end processing, open source by Google.

The major updates for this release include:

• Added support for iOS
• Added optimizer (Adam, AdaGrad, AdaDelta, RMSProp)
• New  math.cos,  math.tan,  math.asin,  math.acos,  math.atan,  math.sinh,  math.cosh, math.abs
• New  math.oneHot and  math.multinomial #160
• Added  NDArray.getValuesAsync() to download values #146 from the GPU asynchronously
• Added math.concat[1-4]D and  added math.slice[1-4]D
• Fix  NDArray.randTruncatedNormal Bug  #172
• Added when the glsl fragment shader can not compile when highlighted

  

• Faster ops restore (min / max / reduce_sum / logsumexp)  #145
• Faster matrix multiplication and convolution/pool  #129
• Added script for model builder training data preprocessing #136
• Reference measurement improvement #140

For more information, please refer to the release notes

Download

• deeplearn.js
• deeplearn.min.js
• Source code (zip)
• Source code (tar.gz)