Kyra is passionate about world-changing tech and sustainability, and happiest when these come together (looking at you, Sama). Stereotypical Dutch, she enjoys urban cycling and eating stroopwafels.
January 22, 2021
11 Minute Read
Large developments in AI and Machine Learning are normally announced as part of a campaign or during important keynote presentations, but we thought we would try and get a sneak peak into 2021 developments in the field. We partnered with our friends from RE•WORK by asking a range of experts in Machine Learning what they believe will be the next big thing.
Machine Learning Software Engineer, Google
Tushar’s predictions for 2021 are very much based on the outcome of work already underway. The former General Motors trailblazer suggested that he was on the lookout for the following in 2021: 1. The “rapid” training, evaluation and productionization for the large sequence models (such as GPT-3, T5). 2. The research towards developing giant ML models that can cater to multi-model inputs from various tasks and that can generalize well to the new tasks.
Director, AI, Pfizer
With over 17 years at Pfizer working in AI, Sean has overseen tasks ranging from quantification of MRIs to multivariate analysis of disease progress and most recently, a global effort in drug safety assessments. When asked what he thought might be on the horizon for this year, Sean suggested “We may train the machine to predict the Machine Learning in 2021.” This rather funny response, if true, could see a move toward the automation in ML we have discussed for some years.
Data Science & Machine Learning, Apple
Alireza’s predictions for the year ahead were quite something, and we love the sound of this! “2021 will be the year when we start to see more tangible ML-driven products; from significant improvements in self-driving cars to a wave of new products based on augmented reality”.
Senior Machine Learning Engineer, Bose Corporation
Working across NLP, Deep Learning and now ML, Shuo has a range of experience. When asked, Shuo suggested that while the question in itself is extremely broad, “I'd say that less dependence on human supervision by leveraging the intrinsic structures of large data is a path ML research is heading toward.” A step closer to the infamous automation buzzword.
Software Engineering Manager, Facebook
Jason’s experience has spanned creative positions at Apple, Google and now Facebook, where he is working on a scalable reinforcement learning platform. When we asked Jason what he thought we could see in ML this year, he suggested “I predict that decision making AI will surge in popularity. Imagine an AI system that can tune constants in your code, or one that monitors your diet and suggests food. We have done a lot in the realm of signal processing but decision making continues to be an area with a lot of engineering trial-and-error.”
Senior Director of AI, CVS
“In 2021, researchers would continue to take a leap in developing complex AI applications related to natural language understanding and generation by leveraging the ongoing advancements in multi-modal (text, speech, image) data fusion-based algorithms and more efficient transformer architectures. In addition, powerful synthetic data generation and augmentation techniques would enhance effective training of machine learning models by alleviating the challenges related to accessing sufficient ground-truth data. We would also see further emphasis on AI model’s ethicality, generalizability, explainability, and reproducibility along with efficient model Ops for a beneficial and robust on-device deployment. Furthermore, with the ongoing proliferation of healthcare data, the AI community would continue to build novel machine learning-based healthcare applications leading to increased AI adoption for providing meaningful decision-making support to end-users.”
Data Scientist, Gartner
“As more and more companies are heading towards AI/ML Maturity and working with more production scale ML/DL deployments, their focus in 2021 will be more on MLOps where they need to work on key aspects of processes like - Model/Data Drift Analysis, Model Interpretability, Data Governance, Model Scaling & AutoML. This year we will finally see many maturity models and practices for ML/DL operations coming up from different vendors and companies which will be very crucial in the long run.”
Senior Data Scientist, Sky
The Senior Data Scientist at Sky when asked his predictions for 2021 stayed very much on the hot topic of COVID. “The goal for 2021 is that Data can support governments beyond counting Covid19 cases and death rates. Given the recent events ML will have to improve massively in inference and causality issues. Forecasting methods using AI do not deliver anything on our historic events scenarios mainly because of the lack of explainability of the generated features.”
Machine Learning Lead, HSBC
2021 will be the Year of MLOps. Many cloud services such as GCP's AI Platform, AWS's Sagemaker have matured over the past years and are readily adaptable by ML teams. In 2021 and beyond Data Scientists would have to work collaboratively with DEs/MLEs to take their beautiful models from lab to production and realize business value.
NLP Research Scientist, Staff ML Team, Stanford
The former Senior ML practitioner at UBER, currently makes up part of Professor Chris Ré's Hazy Research group on ML systems in California. Piero suggested that in industry he believes “models will start to come closer instead of being in separate silos like it happens today in most cases. In academia, I believe there will be more research focused on robustness and generalization.”
Research Assistant, University of Florida
“The year 2020 has significantly changed our lives in many ways. The way we work, communicate and learn has been impacted in many significant ways. Artificial Intelligence continues to be a key technology trend when it comes to the things that will change how we live, work, and play in the near future. I predict that artificial intelligence will be the main technology in the following areas: If any threat such as COVID were to occur, then that should be detected ahead of time. To help businesses prepare for situations like this pandemic, we must understand how users behave during such an emergency.”
Senior Machine Learning Engineer, Autodesk
"Over the last year, we've seen a lot of local AI startups being sold/acquired. Many are struggling to bring AI products to market. I believe the pandemic merely sped up the inevitable. The silver lining to me here is that we've managed to cut down on the hype (noise) giving us time and resources to focus on impact with tangible applications (signal). That's why I think 2021 will be the year we will start to see AI/ML features in production for real - i.e., AI will successfully make the transition from being a spice to a core ingredient."
Senior Computer Vision Engineer, Walmart
“I think 2020 has changed the world around us in many ways, and the retail sector is no exception. The pandemic has taught everyone new ways of shopping whether it's online, curbside pickup or concierge services for senior members. And AI/ML has played a major role in shaping all these solutions everywhere. Going in 2021, I feel rather than quantity of data, the market will shift towards the quality of data being captured. As more and more data points are getting captured to make better ML models; need of the hour is to make light weight models that can run on edge devices to filter and capture just the meaningful data. I call it -- "AI/ML for Intelligent Data collection.”
Research Assistant, Ryerson University
Shaina, a former lecturer and now Research Assistant in ML gave one of the more exciting predictions on the list, which would mean large scale change and positive development in the field in the months to come. “Machine Learning is going to be easier, cheaper and beyond the limitations. The exponential power of the computing resources will be in the hands of laymen, and we see the revolution of the world much more earlier than expected.”
PhD Computer Science, University of Buffalo
Yaman’s research in ML revolves around Adversarial Networks, Interpretability, QA and Speech and is supported by Google doctoral Fellowship. Yaman simply suggested three compartments of ML which he believes will develop rapidly in 2021. “Unsupervised Learning, Explainability, and Fairness in ML.” Surprisingly, it was the only mention of unsupervised learning, which has seen to become somewhat of a buzzword at the backend of 2020.
Chief AI Scientist at AI Systems & Strategy Lab
I feel that the role of knowledge will be increasing. Structures, various knowledge representation types will be essential for the next wave of AI innovation. The lawmakers might soon require AI, ML models to be compliant with multiple statutes or regulations.
2021 is set to be uncertain in many ways. However, as with last year, hard work behind the scenes in AI and ML continues. The range of optimistic yet realistic predictions above, if accurate, could see 2021 as a year in which some of the largest steps in AI and Machine Learning advancement for some years take place.
What do you think could be developed this year?