Ruijing Yang is an AI expert specialized in computer vision, with a strong academic background. At Silo AI, Ruijing works as an AI Scientist, focusing on creating computer vision and deep learning-powered AI solutions with our clients. In this interview, we get to know Ruijing and her work better.
Six years in computer vision
With more than six years of experience in computer vision, Ruijing has both a profound theoretical training, and practical experience on different computer vision topics, such as face detection, facial expression recognition, and object detection. Her work includes mainly dealing with image or video data, where she focuses on any visible objects, such as identifying facial expressions on a human face or objects like cars at traffic intersections.
Prior to joining Silo AI, Ruijing worked as a PhD researcher in Computer Application Technology at Northwest University, Xi’an, China, and as a visiting researcher at the Center for Machine Vision and Signal Analysis at University of Oulu, Finland, with a focus on human facial behavior analysis. Her previous projects include creating computer vision for hospital and home caring, leveraging multiple modalities such as video, audio, as well as physiological signals, and designing and developing protocols for interactive emotion recognition.
With your strong research background, what is different in creating computer vision solutions for “real world” problems?
– There is a big difference between doing research and developing methods for industry problems, though the working pattern remains almost the same: reading the latest papers, coding with Python in Pytorch and Tensorflow frameworks, finding solutions for the problems met during the process. However, the mode of thinking changes dramatically: when creating AI solutions for real world problems, we need to think from our clients’ perspective, and clearly understand their needs and goals.
Another thing that is different between academia and industry, is that competing for state-of-the-art performance is no longer the golden standard. What matters is how well we can solve our clients’ problems with reliable solutions, and that we keep to the project timeline. Usually, we need to consider making improvements in different ways, such as by being more efficient or more intelligent, and by saving in costs. In such cases, the state-of-the-art method may not always be the best solution.
For you as an AI Scientist, data matters a lot. There is a big gap between publicly available datasets that are used in academia and the real world domain-specific data that clients provide. Could you describe these differences?
– First of all, most of the public datasets are collected under certain conditions, and for the general purpose of testing different algorithms.
Take the face detection benchmark data set WIDER FACE as an example. Even though some of the images consider wild scenarios with more variations such as heavy occlusion of the face, low resolution and extreme poses, it’s still quite limited in terms of size and complexity when it comes to real world cases.
For instance, WIDER FACE is considered a relatively large dataset (around 30,000 images with roughly four million labeled faces) that could be used as training data for face detectors in academia. However, in some client cases the size of data is in the magnitude of millions of images with the majority of the images are from hard-to-analyze scenarios.
Second, many of the currently available state-of-the-art models will most likely not work on the specific use case our client has. Therefore, both analyzing domain-specific data and designing the AI model is necessary for client work. In some cases, analyzing domain-specific data involves a significant amount of data collecting and pre-processing first.
Finally, we should always keep in mind that the reliability of the AI model and data privacy are the two vital factors we need to consider before participating in any project. Reliability is crucial to all the machine learning models, but it is of utmost importance in specific domains, such as healthcare. What comes to privacy, many companies are still learning how to best manage their data and privacy. We need to be able to consider privacy risks, build trust with our clients and make us step ahead of other companies in this on-going AI wave.
What do you enjoy in your work at Silo AI?
– I’m always excited about using existing technology to solve real-world problems – this is the mission of technology. One of the best things in working at Silo AI has been the opportunity to work on various interesting projects, as well as to work in different project teams and clients, meeting people from different backgrounds.
Silo AI has a very flexible and warm atmosphere, which I think has made my life much easier. Everything is easy and negotiable, AI experts have great support from our People team, Operational team and Sales team. What matters is that AI experts get to do our best in projects and the rest will be arranged.
Besides working, what do you do?
– I have a deep love for running. Running is my way of meditation. I used to run a half-marathon every year, but I’m taking a break this year due to the pandemic. Meanwhile, I would also like to do strength training in the gym. Running and strength training not only bring me a strong body, but also bring lots of pleasure mentally, with dopamine! Another thing to mention here in Finland is metal music – it gives me a lot of energy (my favorite bands include Nightwish, Children of Bodom, Iron Maiden). It’s so great that Finland is the home of metal music. I love it!
What’s your favorite Silo AI value?
– My favorite value of our company is ‘keep learning’, and now I would like to add ‘build bonds’ too. Learning is a lifelong process. I love to learn new things from different aspects, from technical parts, from projects, from other great people, or from my own or somebody’s mistakes. We humans are social animals. Hence building bonds between clients, between colleagues and exchange ideas are important, communication enriches the understanding of each other and the business models as well.
Thanks for the great chat Ruijing! If you would like to work with a stellar computer vision expert like Ruijing, get in touch with our Talent Acquisition Lead Jenni Kivikoski at email@example.com or check out our open positions at silo.ai/careers.