Hot subtopics in AI research

If you are reading this article, you are already surrounded by AI-powered tech more than you can imagine. From the website in front of you to reading CT scans, AI applications are inevitable.

Generally when people hear about AI they often equate it to Machine Learning and Deep Learning, but they are just two of the many subtopics in AI research. These two are arguably most effective themes in today’s AI world but there are many other subtopics that have gained significant traction in the AI community for their applications and the future potential. In this article we will discuss some of the hot subtopics in the AI research, many of these topics are interlinked and come under broad umbrella of artificial intelligence:

• Large scale Machine Learning
  Machine Learning (ML) is concerned about developing systems that improve their performance with experience. In the last decade progress in AI can easily be attributed to the advances in ML. ML is so popular that it has become synonymous with AI. The researchers are now focusing on scaling the state-of-the-art ML algorithms to large datasets. For more information on ML read this introductory blog.
• Deep Learning
  A subset of ML, Deep Learning (DL) is re-branding of neural networks- a class of models inspired by biological neurons in our brain. DL has been driving force for lots of applications in AI like object recognition, speech, language translation, playing computer games and controlling self driving cars. For more information on DL read this introductory blog.
• Reinforcement Learning
  Reinforcement Learning (RL) is the closed form of learning to the way a human being learns. It consists of an intelligent agent that interacts with its environment smartly to reap a numerical reward. The goal of the agent is to learn sequential actions so as to maximize the long time reward. Like a human being who learns from his experience with the real world, keep exploring new things and updating his values and beliefs, the RL agents works on the similar principle to maximize his own rewards in the long run. In 2017, Google’s AlphaGo computer program used RL to beat the world champion in the game of Go. For more information on RL read this blog.
• Robotics
  Technically speaking, Robotics is a separate branch of its own but it do has some overlap with AI. AI has made robot navigation in dynamic environment possible. How do you make sure that a self driving car goes from point A to point B without harming itself and anyone else in the least time? Advances in DL, RL probably have answers to such questions in Robotics. For more information on robotics read this blog on AI powered Robotics and watch these demonstration videos: 1, 2, 3.
• Computer Vision
  If We Want Machines to Think, We Need to Teach Them to See. Fei-Fei Li, Director of Stanford AI lab
  Computer vision (CV) is concerned with how the computer visually perceive the world around it. Ironically, computers are good at doing mammoth tasks like finding tenth-root of a 100 digit number but struggle in simple tasks like recognizing and differentiating objects. Recent advances in DL and availability of labelled datasets and high computing power have made possible for CV systems to outperform their human counterparts for some of the narrowly defined tasks like visual object classification. For more information on CV read this blog.
• Natural Language Processing
  Natural Language Processing (NLP) is concerned with systems that are able to perceive and understand spoken human language. It consists of sub tasks like speech recognition, natural language understanding, generation and translation. With multiple languages used across the globe, NLP systems could be a real changer. Current NLP research includes developing chat bots that can dynamically interact with humans. For more information on NLP read this introductory blog.
• Recommender Systems
  From what to read, what to buy, to whom to date, Recommender Systems (RS) are everywhere and have completely replaced the annoying salesman in the virtual world. Companies like Netflix and Amazon heavy rely on RS. A RS takes into consideration a user’s past preferences, preferences of its peers and trends to make an effective recommendation. For more information on RS read the following articles: 1, 2.
• Algorithmic Game Theory and Computational Mechanism Design
  Algorithmic game theory considers systems with multiple agents from economics and social science perspective. It sees how these agents make choices in a incentive-based environment. These multi-agent systems can include self-interested human members along with intelligent agents that compete together in a limited resource environment. For more information on this topic you can follow writings of Professor David Parkes. This link is also a good resource.
• Internet of Things
  Internet of Things (IoT) is a concept that daily use physical devices are connected to the internet and can communicate with each other via exchange of data. The data collected could be processed intelligently to make the devices smarter. This article explains how AI could be used to make smarter buildings.
• Neuromorphic Computing
  With rise of Deep Learning that relies on neurons based models, researchers have been developing hardware chips that can directly implement neural network architecture. These chips are designed to mimic the brain at the hardware level. In an ordinary chip, the data is required to be transferred between central processing unit and storage blocks, which results in time overheads and energy consumption. In an neuromorphic chip, data is both processed and stored in the chip in an analog manner and can generate synapses when required, saving time and energy. For more information on development of these brainy chips read these two articles: 1, 2.

Other articles detailing trends in AI research: 0, 1, 2, 3.

References:
Stone, Peter, et al. “Artificial intelligence and life in 2030.” One Hundred Year Study on Artificial Intelligence: Report of the 2015–2016 Study Panel (2016)