Google Pathways architecture claims to solve the limitations of today’s AI
Alphabet Inc’s Google announced the introduction of Pathways, a new AI solution that combines the capabilities of multiple ML solutions and brings them together on a single AI system.
According to Jeff Dean, SVP-Google Research and Health, Google Senior Fellow and also head of Google AI, ML models are subspecialized in individual tasks and rely on a single form of input. To synthesize them on several levels, Google built Pathways. This solution will allow a single AI system to generalize on millions of tasks, to understand different types of data and with increased efficiency. He explains that the solution is to “move us from the era of single-use models that only recognize models to one in which more versatile intelligent systems reflect a deeper understanding of our world and can adapt. to new needs ”.
Register for Analytics Olympiad 2021>>
Dean says Pathways is a solution to the three limitations of current AI models.
– AI models are usually trained to do one thing.
– AI models mainly focus on one direction.
– AI models are dense and inefficient.
Dean argues that today’s AI systems are trained from scratch for new problems and that the parameters of the mathematical model are initiated with random numbers. Each new model trains from scratch to do just one thing, rather than expanding the existing learning, which makes the process much longer. Their solution paths make it possible to train a single model to do several things. The model can have different capacities and be assembled to perform new and complex tasks. According to him, it is similar to the human brain.
The solution can activate multimodal models that simultaneously encompass vision, hearing and language comprehension. The announcement states that “Pathways could handle more abstract forms of data, helping to find useful models that have eluded human scientists in complex systems such as climate dynamics.”
Next-gen AI consists of a single model that is activated in a “dotted” fashion. This means that small, relevant paths across the network can take over the task rather than the entire system. Such an architecture with greater capacity and a variety of tasks can be fast and much more energy efficient.
This is the second Google AI solution to provide multiple solutions to work together. Earlier this week, Google AI proposed a method called Task Affinity Groupings (TAG) to determine which tasks should be trained together in multitasking neural networks.
Subscribe to our newsletter
Receive the latest updates and relevant offers by sharing your email.
Join our Telegram Group. Be part of an engaging community