The Infosys Labs research blog tracks trends in technology with a focus on applied research in Information and Communication Technology (ICT)

« September 2019 | Main | May 2020 »

March 10, 2020

Transfer Learning - the next frontier

Along with the dawn of the industrial 4.0 revolution, AI and ML has made strong inroads into the current and future of IT landscapes. AI seeks to build smart machines and ML involves the software algorithm that programs these intelligent machines to perform without human intervention. AI and ML technologies are playing an integral role in IT growth strategy and offering transformation business values like large returns, real time insights, efficient decision making and driving efficiency across enterprises.
With the market for deep learning software projected to reach a whopping $935 million in 2025, Machine learning (ML) has already emerged as one of the key elements of global digital transformation. Machine Learning is evolving continuously, however conventional ML and Deep learning algorithms have so far been designed to work in isolation and trained to solve specific tasks. Unlike the human mind that can reuse, cross utilize or transfer knowledge across related activities. As a result, further research in the field of deep learning algorithms taking place to replication this human capability. Transfer Learning - a new AI technique has come up that helps to eliminate the need of models learning anything from the scratch. Transfer Learning is based on similar traits of human learning where new models can be built on existing pre-trained models and reuse the learning from existing model to solve related tasks. In other words, Transfer Learning can be used to transfer the algorithmic logic from one ML model to another. For e.g. a pre-trained model that can detect land-based vehicles, when designed with Transfer Learning capabilities would be able to detect air or water-based transportation as well.
In the words of eminent professor and data scientist, Andrew Ng - "After supervised learning -- Transfer Learning will be the next driver of ML commercial success". The Neural Information Processing Systems (NIPS) 1995 workshop on "Learning to Learn: Knowledge Consolidation and Transfer in Inductive Systems" is believed to have given the initial motivation for research in this field. Since then, terms such as Learning to Learn, Knowledge Consolidation, and Inductive Transfer have been interchangeably used for Transfer Learning. 
The use cases associated with Transfer Learning span across multiple industry lines. TL is extensively used in gaming domain, where the tactics learned in a game can be reapplied to play another game, In the field of medical imaging, TL reduces the time to classify millions of images from different categories to detect any disease. Deep Learning algorithms like Transfer Learning work on historical as well as real time market intelligence for complex decision making, forecasting the performance of markets. 71% of Organizational spend are invested in Machine Learning for Cybersecurity. Natural Language Processing (NLP) to become more nuanced with the progress in the world of ML. Primarily used for converting data into text, natural language generation is a key feature of many deep learning systems - and very useful for the preparation of detailed market summaries or reports Businesses can know their customers better by implementing Sentiment Analysis, e.g. models designed for analysis sentiments in twitter feed can be reused for movie reviews. Multiple startups as well as industry behemoths are doing ground breaking research to replicate such a learning onto their machine learning models and come up with better solutions to business problems. Transfer Learning has brought in a new wave in machine learning by reusing existing algorithms, speeding up the process of predictive analysis. This has a direct impact on reducing cost of investment as well as time to train a model.
As per a recent article published in Forbes, CXOs have touted Machine Learning as the future of business culture. According to McKinsey, machines will possibly 'augment employment by around 5 percent by 2030, as well as improve productivity by about 10 percent'. As per expert predictions for 2020, new AI / ML models and insights will be the key enablers of automation to drive efficiency and increase productivity and Transfer Learning is positioned to contribute to the learning curve of Machine Language in a significant way.