We’re thrilled to welcome Ching-Fong (CF) Su to our Leadership Team as VP of Machine Learning. In this role, CF supports the Hyperscience teams that deliver shorter term ML projects that create value for our customers as well as longer term bets that push the technical frontier in our mission to become the world’s leading automation company.
CF brings over 15 years of R&D experience in the tech industries. He’s led engineering teams in fast-paced start-ups as well as big Internet giants. His expertise includes areas of search ranking, content classification, online advertisement, and data analytics. Most recently, CF was the Head of Machine Learning at Quora, where his teams developed ML applications of recommendation systems, content understanding, and text classification models. Before that, he held technical leadership positions at Polyvore (acquired by Yahoo), Shanda Innovations America, and Yahoo Search and was a senior researcher at the Fujitsu Lab of America. To date, CF’s industry contributions include 14 U.S. patents and more than 20 technical papers.
Amidst record growth and open ML Engineering positions around the globe, we sat down with CF to talk about why he joined Hyperscience, the exciting ML projects his team is tackling, and the opportunity at hand.
Thanks for taking the time, CF! Why did you decide to join Hyperscience when you did?
I have been working on various Machine Learning (ML)-based consumer Internet applications throughout the past 15 years. These projects range from a web search engine to content recommendation systems on social network platforms. Those applications provide an excellent environment for ML development. The objective of the ML models is well-defined, which is usually a binary outcome such as whether a user takes action or not. Also, we have a massive amount of data from the users’ behavior on the products. The data provides input for continuous ML training and iteration. However, outside of these Internet applications, many parts of our society have been underserved by ML technologies. For example, core business processes are ready for a significant productivity boost if ML technologies are broadly adopted.
After learning about Hyperscience and speaking with senior leadership, including Peter, I was excited to help build a great AI company. I have seen many start-ups and large companies paying lip service when they approach ML. By contrast, ML plays an essential role in enhancing our product and driving innovation at Hyperscience. Every day, I become more convinced of the business areas and processes waiting to be disrupted by ML. I knew I could add value to the teams here, supporting them to deliver significant value and business impact. So, I decided to join!
What are you most excited about working on at Hyperscience?
I am excited about the opportunity to deliver impact to our customers and leverage all the tools that are at our disposal. When a company initially migrates its offering from a heuristic-based solution to an ML-based approach based on data, they usually deliver a step-function jump on the performance right away. Over time, the return on investment starts to diminish. It gets more difficult to move the meter and typically involves higher cost. In the domain of business automation, the adoption of ML is still nascent. It’s exciting to see the plentiful areas where ML can add tremendous value to our customers.
The progress of ML research in the community is very favorable to us over the past few years. Our product leverages state-of-the-art ML technologies, including image processing, computer vision, and Natural Language Processing (NLP). All have seen groundbreaking progress in academia and industry, and I am particularly excited about the Transformer-based neural network architecture and the growing trends of pre-trained language models. Many researchers are directing their attention to topics such as few-shot or zero-shot model training and low-resource NLP. These subjects are critical to us, particularly as we tackle the open questions in the technical challenges we’re facing. I look forward to seeing our ML teams collaborate with the community and contribute to state-of-the-art research.
How do you think about building your team?
The role of an ML engineer at Hyperscience is a hybrid one. Our team is a combination of ML scientists and software engineers. All the ML technologies in our product originated from ideas inspired by academic research papers and then developed internally. The capability of conducting research and transforming research ideas into products is critical.
Every ML engineer is also a software engineer; None of our teams deliver only research papers or prototypes. We expect team members to write production-quality codes that are used by some of the world’s largest organizations.
Before Hyperscience, the company I worked at had a lab organization, separated from the software engineering teams. The lab was where ML researchers and data scientists focused on research and publication. The communication chasm between the researchers and software engineers was so wide that they added a third role – research engineer – to bridge the gap. At Hyperscience, we avoid separated ML scientist and software engineer roles. There should be no wall between research and product.
Why should an ML Engineer join Hyperscience?
I could offer three reasons.
- Mastery: We are developing cutting-edge ML technologies around OCR, NLP, and beyond. Many open questions exist in the research communities. If you like the challenges and want to contribute to the state-of-the-art NLP research and development, you will be surrounded by a group of talented and research-minded team members at Hyperscience.
- Autonomy: ML is a crucial pillar of our technical stack and is the core of our business. We are experiencing rapid growth, and we are looking for end-to-end owners to look after many of our ML systems and projects. You will never be a small cog in a big machine here; our ML engineers have a massive amount of impact on the company’s business.
- Purpose: If you are a seasoned ML engineer who has worked on consumer Internet applications and is looking to disrupt a new business or industry, I have a question for you: when was the last time you heard directly from the user or customer of your ML products, “Wow, this is like magic. It’s really cool!” At Hyperscience, we have the unique opportunity to develop ML models for business automation that result in significant productivity improvements and better customer outcomes. It’s incredibly rewarding to see our platform cut down disability claims processing times, helping families get the support they need without waiting for months.
What philosophy do you embrace to ensure innovation?
When building a software product, I believe a hybrid of top-down coordination and bottom-up initiatives will yield the best results. The ultimate goal of innovation in a software development context is product and process improvement, rather than growing the world’s knowledge as in an academic environment. So I think high-level guidance and coordination are essential to avoid fragmented moon-shot attempts that try to boil the ocean.
My first job after getting my Ph.D. was at a corporate research lab. Besides conducting research and publishing papers, I had to complete technology transfers to our product group to demonstrate the merit of my work. However, my engineering partners usually found my research output less relevant and of little value to the product. Years later, I was working as a software engineer at an Internet search engine company. I found myself on the other end of the so-called “technology transfer.” My research colleagues tried hard to transfer their research to the product I was building, but I also found it less relevant and of little value to the product. So I believe innovation needs to happen within the guardrails defined by the business needs, coordinated under a central objective.
Innovation can – and does – come from anyone and anywhere across our teams. No one on our ML teams has the “exclusive” right to research. We have some dedicated resources for long-term research projects, but we don’t have fixed research teams or researcher roles. As a leader, I try to incubate the culture, align the incentive, and provide support so that the team members are intrinsically motivated – rather than asked – to pursue innovation. Our ML engineers constantly read the latest research papers and meet regularly to present and discuss what they’ve learned. It’s a purely grassroots effort led by our engineers that speaks to our passion, knowledge and commitment to innovation.