At Spell, we’re working on making high end computing power more accessible for machine learning and deep learning purposes. Spell is a relatively recent entry to the artificial intelligence technology landscape, and comes at a good time as more people are becoming interested in artificial intelligence and putting it to practical use.
In the last couple decades, great strides have been made in mobile device advancement, Internet e-commerce and other areas of technology innovation. Yet artificial intelligence (AI) still struggles from a few roadblocks that make it harder for people to innovate in this space. AI algorithms often require massive computing power, and putting those algorithms to practical use is often not as easy as it seems. Spell came about to reduce the barrier to entry for those looking to test out machine learning and run deep learning experiments.
A core principle at Spell is about treating the process of building things with machine learning and AI as fundamentally a different type of software engineering. To date, the traditional way of telling computers what to do has been to give them explicit instructions - hand-written software programming. The technology industry has gotten quite good at that, with tooling, engineering experience, frameworks, etc. Now however, with the increasing power of AI, there’s the potential to tell computers what to do in a more loose fashion, where they can learn adaptively to solve the problem at hand.
With enough data and sample sets, along with many training runs, machine learning enables computers to adapt to challenges in novel ways. Looking towards the future, all the novelty, all the new things that we see are going to be based on this second paradigm - of learning algorithms showing examples. And that being the new way that you can create delight, novelty and value with technology.
How Does Spell Facilitate Machine Learning?
Spell realized that to make this transition to a new paradigm, software developers need tools to be able to coordinate large amounts of compute power and data. Spell provides a simple command-line interface that allows one to move computation onto the cloud and reproduce experiments consistently. The command ‘spell run’ followed by the command you want to run executes the program in the cloud environment. A popular expression within the company goes “We’re in the make things easy business.”
Developers can choose what kind of hardware they want it to run on, and interact with the results through logs, monitoring, model serving and other tools. But at its essence it takes the basic experience of running a program in the local environment and makes it just as easy to do in a distributed compute environment. Reproducibility is an important feature in machine learning or deep learning, and Spell also makes it easy to rerun experiments in a reproducible manner. And there’s an opportunity for cost savings in terms of resource costs, because Spell can ensure compute resources are used efficiently and not wasted.
Spell’s core customers span a wide range - healthcare, retail, finance - researchers and academics. The common themes these customers need help with include running Jupyter notebooks in the cloud, managing data, getting access to a wide variety of compute types, and all the other tasks involved in running machine learning experiments. The industry is pretty broad, but the work also spans the gap between research and industry. Historically, new breakthroughs in academic research have had difficulty making it into industry as real-world applications. This gap is starting to narrow however, as advances in platforms and tooling make it easier to bring research solutions to practical use.
In the AI world in the last 10 years, several trends have emerged. The research community has been quickly conquering open problems, and at the same time AI has become a cultural phenomenon, entering the public’s perception and day-to-day life. Furthermore, AI increasingly is being applied to practical applications - used to solve problems we currently have in the software world.
About Spell’s History and Vision
Spell was founded in 2016 with the goal of providing access for everyone to the powerful compute clusters that large resource organizations work on. Spell’s strategy is to have a full end-to-end suite of tools on the same infrastructure. The remote execution capability is the critical feature for developers using the tools. If you want to run a large scale neural network running on 8 GPUs in the cloud, the actual mechanics of that can be pretty daunting. With Spell, the infrastructure to execute programs in the cloud is already in place, lowering the barrier to entry for a developer to conduct an experiment or run training.
Looking forward, the leadership at Spell is excited to see how machine learning grows as a tool to solve problems in the world. It will certainly take time, but it’s inspiring to think how it may help in areas like healthcare, as research gradually becomes applied to real-world machine learning tasks. The tools for conducting machine learning are advancing very quickly, bringing more people into the field and making it an exciting time to be working in this area.
Get a personalized demo of Spell and learn how the platform can streamline your Machine Learning Operations https://spell.ml/demo