Google Upgrades AI Hub
Google Cloud has added collaboration tools to its AI Hub launched earlier this year as a clearinghouse for machine learning pipelines and trained models.
A new homepage provides users with access to a roster of “shared private assets” ranging from AutoML natural language and tables to TensorFlow deep learning images.
The hub was launched in April in response to concerns about managing the growing number of machine learning tools while reducing the amount of redundant AI development. That, the company said this week, would promote greater collaboration for data science and machine learning developers, allowing them to build on each other’s work.
To promote collaboration, the hub eases the process of managing permissions to allow greater sharing of trained ML models, Kubeflow pipelines and accompanying notebooks. Kubeflow serves as a workflow automation tool for running, for example, deep learning tasks on the Kubernetes cluster orchestrator.
The hub’s “advanced sharing” features are designed to allow teams or an entire company to share production-ready AI services.
The hub includes models and pipelines from Nvidia (NASDAQ: NVDA) and other AI developers. Nvidia’s notebook on BERT (Bidirectional Encoder Representations from Transformers) describes how model performance for natural language understanding can be boosted with GPUs, the partners said.
Another uses a Kubeflow pipeline for times-series forecasting via TensorFlow and other tools.
“Since releasing AI Hub, we’ve learned a lot about the challenges our first beta customers face bridging gaps and silos in ML projects,” Google (NASDAQ: GOOGL) noted in a blog post. “These new features are a direct result of these ongoing conversations and aim to make it easier to get started with any ML project by building on the great work of others.”
Google Cloud managers said they would continue integrating AI Hub with other machine learning development efforts, including those not currently running on Google Cloud.