To investigate how the brain links temporal intervals of varying lengths, we first used recurrent neural network (RNN) modeling to replicate the time-interval discrimination task. Building on existing experimental data and simulated neuronal activity, they proposed a unified framework for encoding strategies that bridge intervals of different durations. This framework outlines four potential strategies: scaling, absolute, interval-specific, or a combination thereof. Leveraging this framework, we developed a novel algorithm to quantitatively determine the encoding strategy employed by neural activity at both population and single-neuron levels. Applying this algorithm, we found that both biological and artificial neural networks preferentially adopted a hybrid strategy across two experimental tasks.
Related paper:
Zhou, S., Masmanidis, S. C., & Buonomano, D. V. (2022). Encoding time in neural dynamic regimes with distinct computational tradeoffs. PLOS Computational Biology, 18(3), e1009271.
Address:No.138, medical college road 
Postcode:200032 
Tel/Fax:021-54237056
Emial:itbr@fudan.edu.cn