The Google Brain Team has open-sourced Tensor2Tensor (T2T), a tool designed to allow users to replicate and tinker on deep learning research results in less time and on an ordinary PC.
Deep learning – an artificial intelligence function that makes sense of data without the need of an explicit algorithm – has made speech recognition, object detection and machine translation possible. While research results on deep learning-based systems are available to the public, they are hard to replicate and to experiment on as they require complex processes and top-of-the-line computers.
Łukasz Kaiser, senior research scientist at Google Brain Team, in a blog post wrote that Tensor2Tensor facilitates the creation of state-of-the-art deep learning models by “enabling the exploration of various ideas much faster than previously possible.”
According to Kaiser, with Tensor2Tensor, one can replicate and tinker deep learning research results in just one day in an ordinary PC. “Notably, with T2T you can approach previous state-of-the-art results with a single GPU in one day,” Kaiser said. “Now everyone with a GPU can tinker with great translation models on their own: our GitHub repo has instructions on how to do that.”
The senior research scientist at Google Brain Team added that with Tensor2Tensor, one can develop a single deep learning model that incorporates previously defined deep learning models such as image classification, image captioning, speech recognition and translation.