2017 is the year the machine learning conversation shifts from research papers to product roadmaps. Three developments define the moment.
First, Google’s AlphaGo defeats world champion Ke Jie in May, demonstrating that deep reinforcement learning can now outperform human experts in domains previously thought to require intuition.
Second, Facebook releases PyTorch 0.1, offering a dynamic computation graph that makes experimentation dramatically faster than the static graph approach of TensorFlow. The two frameworks will define the deep learning tooling landscape for years.
Third, AWS SageMaker, Google Cloud ML Engine and Azure Machine Learning all reach meaningful maturity, bringing GPU-backed training and hosted inference within reach of teams without dedicated ML infrastructure. For enterprise software teams, 2017 is the year AI stops being a research question and starts being an engineering problem.