France is hosting the 2015 International Conference on Machine Learning: ICML. It is one of the major events focusing on machine learning.
NVIDIA wants to make its GPU attractive for developers so now it concentrates on machine learning as part of their 2015 Maxwell and Tegra X1 initiatives along with neural networks.
The company has already announced new releases for their software libraries/environments, CUDA, cuDNN, and DIGITS.
The new version of CUDA 7.5, is now available as a release candidate. The update focuses on laying the API groundwork for FP16. The CUDA 7.5 introduces support for FP16 data. At the moment NVIDIA is enabling this feature for developers to take advantage from the reduced memory bandwidth.
The next part of the announcement provides some details on cuDNN 3. CUDA Deep Neural Network library (cuDNN) improves Maxwell GPUs overall performance. Going along with CUDA 7.5 the library supports FP16 data formats in order to utilize memory bandwidth.
Finally, DIGITS is NVIDIA’s higher-level neural network software for general scientists and researchers (as opposed to programmers), offering a more complete neural network training system for users who are not programmers or neural network researchers. Moreover, DIGITS supports neural networks training using multiple GPUs.
The company has not provided any information when the final version of its libraries will be made available.