Fanning the Pytorch with Watson Machine Learning

Source: Deep Learning on Medium

A new Pytorch Cloud Partners project from the IBM Cognitive OpenTech team

Pytorch is an open source deep learning framework developed for the Python programming language by Facebook’s AI research group. It leverages the deep learning library Caffe2 (also developed by Facebook) and is compatible with the Open Neural Network Exchange format (ONNX). Although newer than well-known frameworks like Tensorflow, Pytorch’s ease of use for rapid prototyping has made it very popular among researchers and data scientists.

Hallway to Jim Spohrer’s office, IBM Almaden Lab

As deep learning moves beyond the research lab and into production, cloud platforms have become the preferred solution for both development and deployment of these intensive workloads. In a cooperative effort to accommodate this new demand, IBM and other leading cloud platform providers began partnering with the Pytorch project in late 2018.

Lobby, IBM Almaden Lab

Stealing Fire

For three decades, IBM has demonstrated a commitment to code, content, and community through open source participation. As one of the world’s oldest tech companies, IBM’s institutional character is rooted in a distinct long term and big picture perspective.

Recognizing the need for an open AI champion among industry leaders, Jim Spohrer (Director, Cognitive OpenTech), created the Cognitive OpenTech team within IBM’s Open Technology group, under Todd Moore.

While some were lacking in AI experience, the team’s technical chops and open source community know-how accelerated this new knowledge acquisition. Jim recommended reproducing published AI research and studying methods in top AI Leaderboard submissions. By 2018, the team was actively leading workshops, tutorials, and even contributing upstream to open source deep learning projects like Tensorflow and Pytorch.

IBM COG Mixtapes — Mix #1: Pytorch — Side A: Learning
Jim Spohrer’s office, IBM Almaden Lab
Winnie Tsang and Chin Huang, IBM Silicon Valley Lab
Chris Iijima, Chris Hupman, and Paul Van Eck in their shared office, IBM Silicon Valley Lab

Pytorch in the IBM Cloud

Catherine Diep (Solutions Architect and Performance Engineer) felt the example project (or “Code Pattern” per IBM Developer) should make it easier for Pytorch developers to port their work across different cloud platforms. By walking the developer through the MNIST example set up, she highlights Pytorch integration with the Watson Cloud platform. Anyone can customize the project (available on GitHub), replacing the MNIST example with their own existing workload.

Catherine Diep in her office, IBM Silicon Valley Lab

“The point is, we use a Pytorch workload that can run anywhere. In this case, we show you how to bring your Pytorch project into the IBM environment. I think that’s what the beauty of this [project] is, I don’t create a new MNIST workload, I just use a Pytorch example. If I can bring an example into the environment, you can bring your own workload into it. That’s the point!”

Respecting developers’ time and effort is a top concern for Catherine. “If I develop it and then bring it into Watson ML that means you have to develop something specific to work with Watson ML. If you’ve spent the time to develop this workload and you just need an environment, we’re not changing your workload. We show you how to bring your work into our environment.”

Experienced developers like Catherine are very familiar with speed bumps in cross-team interoperability. “During the time we were doing the tutorial,” Catherine reflected, “the location of the Watson ML data changed. In the new version, they created a directory and then put the data in there. We told them in the future, they need better communication about changes like that.”

Catherine feels these kinds of experiences actually increase her effectiveness as a customer advocate. “We’re only working with the offering manager. We don’t have access to developers. Our experience mimics the customer, we don’t get special access to the teams or anything like that. These are the things that are really good for us to see. If we face it, the customer will face it.”

Don’t Stop Retraining

While learning increases our knowledge, it also increases our awareness of how much we don’t know. Certainty, at best, lies in how we engage with the world, particularly with each other. With the predawn of AI barely peeking over the horizon, Ethical AI and the Future of Work are finding a voice among innovation and monetization. Great heroic moments in history are an accumulation of our collective, day-to-day humanity. As we struggle to widen “our circle of compassion“* amid Homo economicus colonialism, what sort of AI monument are we bricklaying? Perhaps an open source model that recognizes everyone as a stakeholder is one path towards a shared, communal AI.

Coffee Break, IBM Silicon Valley Lab
IBM COG Mixtapes — Mix #1: Pytorch — Side B: Challenges
Chin Huang and Tom Truong in a team meeting, Silicon Valley Lab. Photo Credit: Paul Van Eck

“The world we live in today I can’t tell you what’s going to be the hot thing two years from now. You have to be constantly learning new stuff and looking for ways to sharpen your skills, and grow with the technology.” As a manager, Tom Truong (Director, Cognitive OpenTech & Performance) highlights potential and adaptability. “That’s the amazing thing about the team that I have… I can tell you twelve people that probably have twelve different backgrounds, not peas from the same pod. Very diverse, very different. To me that’s an advantage, an asset. I can tackle a lot of these different projects because I can pick and choose and assemble a team that can do it.”

Ton Ngo’s Office, IBM Silicon Valley Lab

We invite you to try out the project and make it yours. In the spirit of open source, please do share any issues and pull requests!

Listen to the entire mixtape here:

Featuring clips from interviews with Jim Spohrer, Thomas Truong, Catherine Diep, and Chris Ferris.

IBM COG Mixtaps — Mix #1: Pytorch

[*] “A human being is part of the whole, called by us ‘universe,’ a part limited in time and space. He experiences himself, his thoughts and feelings, as something separate from the rest — a kind of optical delusion of consciousness. This delusion is a kind of prison for us, restricting us to our personal desires and to affection for a few persons nearest to us. Our task must be to free ourselves from this prison by widening our circle of compassion to embrace all living creatures and the whole of nature in its beauty.” Quote taken from Dear Professor Einstein: Albert Einstein’s Letters to and from Children, Alice Calaprice (Ed.), Princeton University Press, 2002.