CURRICULUM LEARNING: SEQUENTIAL ACQUISITION OF TASK COMPLEXITY ENHANCES NEURONAL DISCRIMINABILITY

Curriculum Learning (CL) is a learning strategy where tasks of increasing complexity are acquired sequentially. This strategy, commonly used in educational and machine learning contexts, accelerates learning. However, its neural underpinnings remain largely unexplored. We investigated the behavioral and neural mechanisms of CL using a novel auditory Go/No-Go paradigm in mice.
We designed a multi-staged appetitive Go/No-Go learning paradigm where a hard auditory discrimination task was preceded by an easier one. Both tasks required the mice to discriminate between a pulse train with a single tone (No-Go) and a pulse train with alternating tones (Go). The key difference between the tasks was the nature of the tones: the easy task used pure frequency tones (PT), while the hard task used harmonic complex tones (HCT). We compared a cohort undergoing progressive training from an easy to a hard task against control cohorts trained exclusively on PT or HCT. Throughout training, we monitored neuronal activity in the auditory cortex using chronic in vivo electrophysiology.
Our results demonstrate that sequentially structuring tasks significantly enhances both learning speed and accuracy. We found that feature commonality between tasks is a crucial requirement and that the effect may be bidirectional. Electrophysiological recordings revealed that training improves neuronal discriminability; notably, training on the easy task alone enhanced neural discriminability for the subsequent hard task.
These findings elucidate the mechanisms allowing progressively complex tasks to facilitate learning. We propose that CL relies on schema formation, where the brain organizes information efficiently to reduce computational load, thereby optimizing performance.