Optimal prediction with resource constraints using the information bottleneck.

Area of Science:

• Computational Neuroscience
• Information Theory
• Evolutionary Biology

Background:

• Organisms must encode external information for stimulus response, facing resource limitations that necessitate signal compression.
• Predicting future environmental inputs offers significant advantages across various biological systems.
• The information bottleneck (IB) framework quantifies the trade-off between retaining past information and predicting future states.

Purpose of the Study:

• To compute the trade-offs between faithful past representation and future prediction using the IB approach for input dynamics of varying complexity.
• To identify optimal motion representations for accurate prediction and transferability across different contexts.
• To explore the role of long-term memory in internal representations for non-Markovian dynamics and evolutionary population dynamics.

Main Methods:

• Application of the information bottleneck (IB) framework to analyze signal compression and prediction.
• Analysis of input dynamics with varying complexity, including motion prediction and non-Markovian processes.
• Quantification of information transferability for different motion representations.

Main Results:

• Depending on input dynamics, velocity or position information is crucial for accurate motion prediction.
• Specific motion representations exhibit higher transferability for prediction in novel contexts.
• Long-term memory significantly shapes internal representations in non-Markovian dynamics.
• Prediction in evolutionary population dynamics correlates with clustering allele frequencies into distinct memory units.

Conclusions:

• The IB approach effectively models the balance between past representation and future prediction in biological systems.
• Understanding information compression and transfer is key to deciphering efficient biological computation.
• Internal representations are shaped by the need for prediction, with memory playing a critical role in complex dynamics.