Researchers at Ohio State University have made advancements in creating artificial intelligence that learns much like humans. This new method, known as 'continual learning', could champion the development of AI with lifelong, human-like learning capabilities, paving the way for a new era of intelligent machines.
Embracing continual learning in AI
Deep-dive into the concept of 'continual learning', an approach that mimics humans' capacity to build on prior experiences to learn new things. The researchers at Ohio State University gave artificial neural networks, the brains behind AI, a human-like ability to continuously acquire new skills while retaining previously learned knowledge. This method could significantly improve the way AI systems perform and adapt.
Diversity in AI training tasks enhances retention
The study interestingly revealed that diversity is key in training AI. Similar to human learning, artificial networks perform better in retaining information when exposed to diverse, unrelated tasks rather than ones that share common features. Teaching an algorithm to handle varied tasks from the get-go can significantly expand its capacity to absorb and retain new information.
Advancement towards human-like learning in AI
The advancements in continual learning research bring us one step closer to AI that mirrors human-like, lifelong learning capabilities. This could potentially streamline the scaling up of algorithms and their adaptation to ever-changing environments. A giant leap for AI, this could open up endless possibilities and applications in a variety of sectors.
In another notable presentation at the conference, a team from MIT showcased a technique that could potentially thwart the creation of deepfake images. They achieved this by injecting small disruptive pieces of code into the source images. This innovation could prove to be a game-changer in the fight against the misuse of AI in creating deepfakes.
Google, a tech giant and major participant at the conference, announced that its AI and machine learning research is featured in over 80 scientific papers at the ICML program. These papers cover a wide range of topics including 3-D protein modeling with AlphaFold, advances in fusion science, and new models like PaLM-E for robotics and Phenaki for generating video from text. Google continues to lead the charge in AI and machine learning research.