Microsoft Research AI

Big tech companies are often hailed for their tech-based products. But have you ever thought of what they undergo and how much effort they put to unleash such futuristic products? Yes, most of the big tech companies have a separate unit for researchers who engage in artificial intelligence, machine learning, cybersecurity, deep learning, data science, etc researches. Similarly, Microsoft also has a group of scientists working for its Microsoft Research AI initiative. They work to use technology to the fullest and merge it with real-world applications. The Microsoft Research AI team has people from diverse backgrounds and geographical locations, working with a simple motive to make technology better. They cover many aspects of practical technologies including algorithms and theory, computer vision, data platforms and analytics, hardware and devices, programming languages and software engineering, quantum computing, security, cryptography, and technology for emerging markets. In this article, IndustryWired aims to showcase some of Microsoft Research’s recent achievements and discoveries.

A Synthetic-Face Dataset to Combat Discrimination

The most challenging task in a machine learning process is not choosing the right model, but finding the perfect data to train the model. Besides, the data is welcoming unwanted issues on the bias, racism, sexism, etc. Therefore, companies are working to keep a distance between emerging technologies and discriminatory aspects. How about using computer graphics to make a more valid representation of training data? Yes, that is what Microsoft Research AI is up to. They have developed a new method of acquiring training data for faces by rendering 3D face models with an unprecedented level of realism and diversity. 

Using Real-world Disaster Data to Predict Responses

Covid-19 is the biggest example of whatever can happen anytime to the world. Following the catastrophe, many unexpected natural disasters like flash floods, hurricanes, storms, wildfires, and earthquakes have ravaged the people. Therefore, Microsoft AI Research realized that it is time for them to look out for such hazardous events beforehand. They use real-world data to make sense of data structure and inform the real-world responses.

DeepMC, a Tool to Predict Micro-Climate Changes

Although existing technology can predict massive climate changes and upcoming disasters, little do we have use for micro-climate moderations. Microsoft Research AI has developed a tool called DeepMC to predict micro-climates or accumulation of climatic parameters formed around a relatively small region. DeepMC helps agriculture and other relative sector find the micro-climate changes with 90% accuracy. They use IoT sensors in the location to monitor the changes. 

CausalCity, a Simulation Environment to Train AI Algorithm

Despite technology walking alongside us for a long time, artificial intelligence is unable to take over human intelligence. Even in the 21st century, technology is struggling to make rapid-decisions based on circumstances. To address this concern, Microsoft Research has come up with a high-fidelity simulation environment called CausalCity that is designed to develop algorithms, which improve causal discovery and counterfactual reasoning of AI. 

Building Task Bots with SOLOIST

Businesses are increasingly using virtual assistants or chatbots to answer consumer queries. Despite their efforts, there is a gap between customers and machines due to varied communication styles and language. While many tech companies are working to bridge the void by unleashing chatbots that communicate like humans, Microsoft Research AI has come up with SOLOIST (TaSk-Oriented DiaLOg WIth A Single Pre-Trained Model) to ease the burden on building user-friendly task bots. They use transfer learning and machine learning to streamline the process.