My Takeaways From Intel® AI DevCon

Intel Artificial Intelligence Developer Conference (AIDC) was a two-day conference that took place at the San Francisco Palace of Fine Arts on May 23–24, 2018. Unlike most conferences held by tech companies, this conference was highly technical and low level in which scientist from Intel and her partners showcased their latest developments in Artificial Intelligence. Technical hands on labs gave the opportunity to the participants to implement state of the art machine learning and deep learning methods. A great deal of the talks and labs were centered on using Intel technology such as Intel® Xeon® processors and Cloud Technology for developing machine learning models for training and inference purposes.

It is amazing to see how much emphasize Intel is putting on Artificial Intelligence (AI) as a game changer. Intel is putting tremendous amount of energy and resources is all aspects of AI including hardware, software, developer, and data. In the past two years, Intel created programs like AI Student Ambassador that provides free resources such as computing power, hardware, and travel sponsorship to student developers across the globe. I was lucky enough to be part of this incredible program and had the opportunity to brainstorm with student developers from countries such as India, Nepal, Poland, England, United States, Israel, Egypt, Kenya, and South Africa.

Here I list some of my takeaways from Intel AIDC.

• Intel Xeon processor has a pretty good performance during the inference process for machine learning systems and it is even competitive to GPUs in terms of time and consumes less power.
• Intel provides cloud-computing services to her customers for machine learning purposes. One of the perks of using these cloud services is that Intel provides software optimization as part of the service for her customers so that they can use their resources such as Intel Xeon processor more effectively.
• Clarifai* is a very cool AI solution startup based in New York that uses AI technology for suggesting items to the users based on the user input image. Suppose you have a picture of a friend and you really like her cloth but how can you tell a machine learning system to specifically look for the cloth? Clarifai allows its users to pick specific parts of the picture and then finds similar items on its database. As another example, users can give their own input images to the system in order to train the system to recognize similar objects.
• Intel utilizes Uber Horovod* distributed training framework in its cloud platform in order to make distributed training fast and easy.
• Naveen Rao, Intel’s VP of AI product group, talked about an Intel software technology called Ngraph. Ngraph is an open source library for developing frameworks that can efficiently and seamlessly run deep learning computations on a variety of compute platforms including different types of CPUs and GPUs.
• Intel Nervana Systems provides the following three cool GitHub* packages that give power to developers in areas such as natural language processing (NLP), Neural network compression, and reinforcement learning.

NLP Architect: NLP Architect by Intel AI Lab is a Python* library for exploring the state-of-the-art deep learning topologies and techniques for natural language processing and natural language understanding

Neural Network Distiller: A Python package for neural network compression research.

Reinforcement Learning Coach: Reinforcement Learning Coach by Intel® AI Lab enables easy experimentation with state of the art Reinforcement Learning algorithms

• If you watched Winter 2018 Olympics opening, Intel uses Swarm of Drones to create different shapes in the sky. If haven’t watched it, I highly recommend watching this video. Now, Intel wants to take this to another level for the Summer 2020 Tokyo Olympics. Here is the link to the Intel challenge for the best idea.

Andrew Ng, Stanford Professor who founded Coursera, delivered the closing keynote speech. There were a lot of excitements in the crowd since most of the people took his online machine-learning course. He gave some a introduction about AI and talked about some of the trends in AI. One aspect of his talk that grabbed my attention was the roadmap that he put forward for the future of AI. In particular he listed the followings as the roadmap that could help developers in creating the next generation of AI products.

• Strategic Data Acquisition since data is one of the major building blocks of any AI system. Sometimes in machine learning we start we small amount of data which is not enough. However, with this small amount of data we can build a model that can annotate some new datasets. The generated data can be added to the original data. Although the process is not as easy as I describe and it needs some level of human supervision.
• Unified data warehouses so that developers can easily access data in order to test their hypothesis. Plenty of places require a lot of organizational signatures in order for employees to have access to the data. Making this process smoother can facilitate development of AI products.
• AI Strategist as someone who could draw the overall roadmap of AI products that developers are trying to build. AI strategist is someone like product manager but has a deeper understanding of the AI systems and knows how they are supposed to work as a system.

Group photo of Intel student ambassadors with Intel program manager, Niven Singh.

Originally published at software.intel.com on June 6, 2018.