Transfer Learning Sets New Frontier for Machine Learning While Contributing into Cheaper Robots

In present times, technologies are converging their ways to offer more to the existing. A number of institutions are merging the advents of different technologies together to yield better outcomes. Sharing the features and their abilities between each other, they always have capacity to add more value to the business operations. Specially combining with AI, robots can enhance their ability to use previously gained knowledge from one context to another. Also, these robots function same as expensive ones.

Observing the market adoption of robotic technology, we can say that it is still facing challenges in coming out from labs and function smoothly in worldly environment. It is convenient to design robots with expensive sensors and peripherals to set an ideal robotic navigation example in laboratory environment but in real world these examples are expensive and consumer-unfriendly while being messy and imperfect for practical exposure.

To crack this challenge, Vrije Universiteit researchers applied a type of machine learning commonly known as transfer learning. Transfer learning is the process of grasping the information the algorithm has learned in one context and apply it to another. In this context, the process could be adapted to operate a robot in lab and apply the information gained from it to those performing in real world. In other words, robots are capable of training themselves first using better sensors in a better environment and further deploy those learnings even while using cheap sensors in practical world.

To evaluate their observations and idea, the experts developed a robot in a parallel environment where it was supported with eight proximity sensors and a single camera. The researchers found that when the robotic algorithm employed transfer learning to plot decisions (using camera only), it happened to learn faster navigation around the room in comparison to when it used no transfer learning technology.

The entire experiment proves that transfer learning is the new frontier for machine learning, or in broader sense artificial intelligence technology. Accelerated advancements in transfer learning have extracted the insights prerequisite for training good-performance models explicitly. A number of studies and researches going on are directed towards transfer learning offerings and purpose. Latest technologies do need models that can transfer knowledge to new task and easily adapt to new domains.

Additionally, the merger of two technologies always opens up new interesting roads for future work and further developments. It makes the research and technology investigation more evident and worth investing into for absorbing greater benefits. Also, it will be interesting to know how transfer learning is transforming robotics as well as other technology implementations across the multiple industries.