As the boundaries of artificial intelligence (AI) continually expand, researchers grapple with one of the biggest challenges in the field: memory loss. Known as “catastrophic forgetting” in AI terms, this phenomenon severely impedes the progress of machine learning, mimicking the elusive nature of human memories. A team of electrical engineers from The Ohio State University are investigating how continual learning, the ability of a computer to constantly acquire knowledge from a series of tasks, affects the overall performance of AI agents.
Bridging the Gap Between Human and Machine Learning
Ness Shroff, an Ohio Eminent Scholar and Professor of Computer Science and Engineering at The Ohio State University, emphasizes the criticality of overcoming this hurdle. “As automated driving applications or other robotic systems are taught new things, it's important that they don't forget the lessons they've already learned for our safety and theirs,” Shroff said. He continues, “Our research delves into the complexities of continuous learning in these artificial neural networks, and what we found are insights that begin to bridge the gap between how a machine learns and how a human learns.”
Research reveals that, similar to humans, artificial neural networks excel in retaining information when faced with diverse tasks successively rather than tasks with overlapping features. This insight is pivotal in understanding how continual learning can be optimized in machines to closely resemble the cognitive capabilities of humans.
The Role of Task Diversity and Sequence in Machine Learning
The researchers are set to present their findings at the 40th annual International Conference on Machine Learning in Honolulu, Hawaii, a flagship event in the machine learning field. The research brings to light the factors that contribute to the length of time an artificial network retains specific knowledge.
Shroff explains, “To optimize an algorithm's memory, dissimilar tasks should be taught early on in the continual learning process. This method expands the network's capacity for new information and improves its ability to subsequently learn more similar tasks down the line.” Hence, task similarity, positive and negative correlations, and the sequence of learning significantly influence memory retention in machines.
The aim of such dynamic, lifelong learning systems is to escalate the rate at which machine learning algorithms can be scaled up and adapt them to handle evolving environments and unforeseen situations. The ultimate goal is to enable these systems to mirror the learning capabilities of humans.
The research conducted by Shroff and his team, including Ohio State postdoctoral researchers Sen Lin and Peizhong Ju and Professors Yingbin Liang, lays the groundwork for intelligent machines that could adapt and learn akin to humans. “Our work heralds a new era of intelligent machines that can learn and adapt like their human counterparts,” Shroff says, emphasizing the significant impact of this study on our understanding of AI.