A Survey on Probabilistic Models in Human Perception and Machines.

Area of Science:

• Computational neuroscience
• Machine learning
• Signal processing

Background:

• Extracting information from noisy signals is crucial for biological and artificial perception.
• Bayesian probabilistic models are powerful tools in human perception and machine signal processing (SP).
• Limited integration exists between these fields despite analogous problems like noise reduction and source separation.

Purpose of the Study:

• To compare applications of probabilistic models in machine SP and human psychophysics.
• To identify commonalities between models used for brain processes and intelligent machines.
• To foster interdisciplinary convergence between computational neuroscience and artificial intelligence.

Main Methods:

• Review and comparison of selective applications of probabilistic models.
• Focus on audio and audio-visual processing.
• Illustrative examples including speech enhancement, automatic speech recognition, audio-visual cue integration, source separation, and causal inference.

Main Results:

• Probabilistic models offer a unified framework for tackling signal processing challenges.
• Significant overlap exists in the application of Bayesian methods across disciplines.
• Common principles can be identified in how biological and artificial systems process noisy signals.

Conclusions:

• Commonalities in probabilistic models can bridge the gap between human perception and artificial intelligence.
• This interdisciplinary approach can lead to more robust and intelligent systems.
• Further integration of these fields holds significant potential for future research and development.