Bengaluru-based Tata Elxsi has been enabling technology-based innovations over the past 25 years. From self-driving cars to video analytics solutions, it has a wide range of innovations enabled by AI and analytics. The Artificial Intelligence Centre of Excellence (AI CoE) by Tata Elxsi deals with the growing needs for intelligent systems. Its cloud-based integrated data analytics frameworks, with patent-pending technologies, enable customers to quickly implement and configure the landscape to obtain actionable insights and better results.
One of the important offerings by the company is the Cognitive Video Services Framework which is essentially an AI-Based Video Analytics solution that helps in tasks such as personalising content for users, transforming video into value using AI, suggesting new revenue generation, automating the content analysis, and more.
Analytics India Magazine got in touch with Biswajit Biswas, Chief Data Scientist at Tata Elxsi to further understand some of the projects they are working on, how AI CoE addresses the growing needs of intelligent systems, AI in video analytics and more.
What are some of the key AI projects that Tata Elxsi is currently working on?
Some of the key projects we working on now are:
- AI for the Media industry for content analysis, recommendation, ad analytics, predictive analytics for fault-tolerant systems and improve NPS score
- Building smart appliances with a rich blend of AI and UX
- Autonomous navigation systems software for different types of transport systems
- AI for edge devices complimenting IoT and Industry 4.0 needs
- AI for situational awareness and operator environment analysis
- State-of-the-art conversational interfaces for various applications ranging from medical chatbots, support services in telecom, cars and smart appliances
Tell us about Tata Elxsi’s Artificial Intelligence Centre of Excellence (AI CoE). How is it addressing the growing needs of intelligent systems?
We work on various areas of AI and machine learning, and the problem statement varies from a simple everyday use-case where AI can enhance the quality of everyday life to advanced autonomous driving. We use AI to make things smart and intuitive, for example, a refrigerator which can track the usage and optimise energy; a microwave which ‘knows’ how to process the content; an intelligent and autonomous washing machine; intelligent robots with ‘eyes and ears’ which can be used right from healthcare to housekeeping.
We are developing state-of-the-art AI solutions for the media industry. Content curation, moderation, generating watchable suiting various profile groups, the recommendation of content, the recommendation for Ad insertion are some of the areas we have been delivering AI solutions to our customers.
For industrial use-cases, we are taking automation to the next level where machines know when there is something wrong and ask for preventive maintenance before things turn bad.
Some of the work we feel good and take pride in is our predictive analytics and anomaly detection capabilities. We got the predictive analytics part right the first time early on, where we could predict network failure a few hours before it actually could happen. This is real-time processing, and all we have is a few minutes window to predict and pinpoint which node would go wrong so that we can give enough room for action to be taken.
We also take care of making the AI model ‘explainable’. Much of our work is to find the root cause of an anomaly or error and help solve them either using AI or otherwise. In some cases, the error is rare but very costly if it occurs. Hence, we need to drill down to the root cause using the available data. We work on a set of machine learning models which specialise in rare sequence extraction and construct the time series of correlated patterns which can decipher the cause of the error or anomaly. This is helpful for debugging and is a very time-consuming process if done manually. With AI, we provide this insight at scale using advanced ML algorithms we have worked on.
How has AI helped personalise content for users? Please explain in detail how AI helps here.
AI has brought about ‘extreme personalisation’. Apart from understanding megatrends and patterns in the industry, AI also can be equally good at providing personalised recommendations with high accuracy. It takes micro details into account, for example, programmes being watched, how long are they being watched for, which keys are pressed, search preferences in the web and more. Every detail is analysed and matched, and the recommendation is continuously updated.
This is possible because AI algorithms are becoming more accurate, and better-curated data are being fed. Also, the way AI models are operating on a large scale, where several hundred or even thousands of them are working in tandem in more autonomous ways.
Which sectors is this offering applicable in and how?
‘Personalisation’ is applicable in almost all use cases we are working in. It need not be restricted to just movie recommendation or choice of restaurant or gourmet cuisines. Personalisation is applicable in other areas such as ensuring optimum cabin comfort in car, choice of driving parameters, autonomous driving, health monitoring, customer care, help desk automation, and more. We are working with the state-of-the-art implementation of AI-based personalisation in all the areas we mentioned above.
How is AI aiding automated content analysis? How is Tata Elxsi enabling it?
Automated content analysis is needed for the hour given the volume of content being generated, including UGC (user-generated content). AI is rightly placed to help and make sure the right content is served to the right users at the right time. Content analysis using AI involves analysing the video, audio, music, speech for generating a rich set of metadata to further downstream processing. Tata Elxsi has built a Video cognition platform called AIVA to do all types of Video cognition by generating rich metadata by performing video analytics.
Tata Elxsi AIVA Cognitive Video Platform is an intelligent video analytics service aimed at providing a single generalised solution across diverse domains. With AI-based visual cognition platforms, most of the mundane tasks can be automated to deliver enhanced workflows, improve customer experience & productivity, optimise processes and create newer avenues for monetisation.
What are the various use cases where the company’s AI solutions have helped OTT platforms?
Below are some use-cases which Tata Elxsi has implemented specifically to OTT services using AI:
- Recommendation engine
- Scene recognition
- Automated subtitle generation
- X-ray feature (actor, scene recognition)
- Content metrics for Ad Analytics
- Fully automated content moderation (screening objectionable content)
- Sports highlights the generation
- Video summerisation for content sharing (highlight emotive and high energy scene content)
What are the challenges of designing AI models for video content?
Some of the challenges of designing AI models for video content are :
- Video contents are very dynamic in nature, with a lot of variations at the scene level, colour composition, lighting effect, image resolution and so on
- There are also synthetically generated content (like cartoon characters, 3d animations) which needs a different type of processing by AI layer
- Distinguishing fake content, copyright violated content
- Variation of languages and dialect when comparing audio metadata along with video metadata
- Some of the description for NSFW (unsafe user-generated content) is pretty vague, like hate-mongering content which depends on subjective interpretation and overall context of the scene
- Compute power to process video in real-time for hundreds of channel simultaneously
What are some of the analytics/AI tools used in the front-end and back-end?
We have developed proprietary tools AIVA for Video analytics and SymanTEx for speech and audio analytics targeted to the media industry. Together these two tools can support various use cases as mentioned above. These tools employ hundreds of deep learning models which work in tandem along with other subsystems, coupled with video and image pre- and post-processing tools, topped with a set of business logic to address industry-specific needs. The focus of these AI tools are to provide high automation for processing media content and deliver to the targeted system with a ‘zero-touch’ process.