The demand for AI in the workplace is unquenchable, but the problem is in developing and maintaining the necessary support infrastructure. According to a 2020 IDC poll, a lack of data to train AI and low-quality data, as well as data security, governance, performance, and latency challenges, remain important impediments to implementation. In fact, a third of the companies polled say they spend a third of their AI lifecycle time on data integration and preparation rather than data science.
Josh Tobin was a researcher working at the interface of robotics and machine learning. His research focused on robotic perception and control challenges using deep reinforcement learning, generative models, and synthetic data.
He also co-organizes Full Stack Deep Learning, a machine learning training programme for engineers to learn about production-ready deep learning. Josh earned his PhD in Computer Science from UC Berkeley, where he was advised by Pieter Abbeel and worked as a research scientist at OpenAI for three years. Finally, Josh made this fantastic field guide on deep neural network debugging.
Tobin and Vicki Cheung, who previously led infrastructure at OpenAI and was a founding engineer at Duolingo, are the co-founders of Gantry. This service promises to assist AI development teams in determining when and how to retrain their AI systems. While teaching a deep learning course with Vicki Cheung at UC Berkeley in 2019, Tobin noticed the pattern firsthand. He and Cheung identified an inflexion point in AI's history: firms had been investing in AI for the preceding ten years to keep up with tech advancements or to assist with analytics. Despite some vendors proclaiming the "democratization of AI," most enterprises have found it challenging to develop AI-powered products.
Tobin, in an interview, said, "The main challenge in building or adopting infrastructure for machine learning is that the field moves incredibly quickly. For example, natural language processing was considered out of reach for industrial applications just a few years ago but is rapidly becoming commonplace today,". He continued, "That's why we're building a continuous machine learning improvement platform."
Tobin states that Gantry can summarise and visualize data during the training, evaluation, and deployment stages since it interfaces to existing apps, data labelling services, and data storage.
Gantry has raised $28.3 million from a $23.9 million Series A round and a previously unknown $4.4 million seed round. The Series A was led by Amplify and Coatue, with participation from OpenAI president and co-founder Greg Brockman and Pieter Abbeel, co-founder of industrial robotics startup Covariant.
Ingesting datasets and understanding the associations between distinct data points within those sets is how AI systems learn to generate predictions. However, AI systems are prone to failure in the actual world due to the fact that real-world data is rarely static; therefore the training set is rarely representative of the real world for long.
Tobin and Cheung believe Gantry's "continual" learning system — infrastructure that can adjust a system to a constantly changing stream of data — holds the answer. Tobin explained that Gantry is meant to provide a single source of truth for AI system performance, allowing users to see how the system is performing and how to improve it by using workflow tools to specify metrics and the data slices on which to compute them.
Gantry belongs to the MLOps (machine learning operations) software category, which aims to streamline the AI system lifecycle by automating and standardizing development activities. Cognilytica, an analytics startup, expects that the global market for MLOps solutions will be worth $4 billion by 2025, up from $350 million in 2019. This is due to the increasing adoption of AI.
According to Tobin, "The days of poor enterprise customer experience are over — customers now expect an experience that is as seamless, consistent and intuitive as what they've come to expect from modern tech companies. Machine learning makes it possible to deliver these experiences at scale. However, machine learning-powered products are expensive to build and pose brand and customer experience risk because models can fail in unexpected and harmful ways when they interact with users,". He continues, "Gantry helps enterprises develop seamless machine learning-powered customer experiences with less risk and lower cost by providing infrastructure and controls required to safely maintain and iterate on their machine learning-powered product features."
Tobin also said that the funding would be put, in part, toward customer acquisition, in addition to expanding the size of Gantry's 22-person team.
