Implications of Information Theory for Computational Modeling of Schizophrenia

Steven M. Silverstein, Michael Wibral, William A. PhillipsComputational 2017 · DOI

TL;DR

This paper proposes that information theory, a framework for quantifying information and uncertainty, offers a powerful lens for understanding the cognitive and neural deficits in schizophrenia. By applying concepts like entropy, data compression, and more advanced ideas like infomax and coherent infomax, the authors suggest a new way to model and potentially treat the disorder's complex symptoms.

What Problem Does This Paper Address?

Schizophrenia is a complex and debilitating psychiatric disorder characterized by a wide range of symptoms, from perceptual distortions to disorganized thinking. Despite extensive research, a cohesive theoretical framework to explain the underlying mechanisms remains elusive, hindering progress in treatment. The authors argue that existing computational models, while useful, often focus on specific brain regions or correlations rather than the fundamental information processing disruptions that might underlie these diverse symptoms. They aim to bridge this gap by demonstrating how information theory can provide a unified, mathematically precise framework for understanding these deficits.

Methods

The paper systematically introduces concepts from information theory and applies them to schizophrenia. It starts with foundational ideas:

  • Shannon Information and Entropy: Quantifying uncertainty and the reduction of uncertainty by a message.
  • Data Compression and Block Coding: Strategies the brain might use to efficiently process information, analogous to algorithms like Huffman coding.
  • Signal-to-Noise Ratio Enhancement: How to improve information transmission in noisy systems, relevant to the increased noise observed in schizophrenia.

It then moves to more advanced concepts:

  • Infomax: The principle of maximizing information transmitted through a system, even with data reduction.
  • Coherent Infomax: An extension that prioritizes currently relevant information by using contextual inputs to modulate processing.
  • Partial Information Decomposition (PID): A framework to break down information into unique, shared, and synergistic components, allowing for a more nuanced understanding of how different inputs contribute to an output.
  • Coding with Synergy (CWS): A goal function where joint information from multiple inputs is essential for determining the output, crucial for tasks like mismatch detection.
  • Predictive Coding: The brain's ability to predict incoming sensory information and signal prediction errors.

The authors use these theoretical tools to reconceptualize various cognitive impairments in schizophrenia, such as slowed processing, reduced attention, sensory gating deficits, perceptual and thought organization issues, and cognitive control failures.

Datasets

Not discussed in the paper. The paper is a theoretical review and proposal for a modeling framework, not an empirical study reporting on specific datasets.

Key Insights and Ideas

  • Schizophrenia as an Information Processing Disorder: The core idea is that many symptoms of schizophrenia can be understood as failures in efficient and accurate information processing, rather than just deficits in capacity or speed.
  • Noise as a Central Factor: Increased neuronal noise in schizophrenia can lead to compensatory strategies that, while adaptive, might manifest as symptoms like hyperintense perception or disorganized thought.
  • Context is Key: Advanced information theory concepts like coherent infomax highlight the brain's need to integrate contextual information with sensory input. Failures in this integration are strongly implicated in schizophrenia.
  • Unified Framework for Diverse Symptoms: Information theory offers a potential unifying framework for understanding seemingly disparate symptoms, from perceptual anomalies to thought disorder, by linking them to common underlying information processing disruptions.
  • Neurobiological Plausibility: The authors connect information theory concepts to emerging neurobiological findings, such as distinct sites for processing driving and contextual inputs in pyramidal neurons, suggesting a biological basis for these computational models.

Key Results

This paper is primarily a theoretical review and conceptual framework proposal. Therefore, it does not present quantitative results from experiments or simulations. Instead, its "results" are conceptual:

  • Novel Explanations for Cognitive Impairments: The paper provides information-theoretic explanations for phenomena like slowed processing, reduced attentional capacity, and sensory gating deficits.
  • Reconceptualization of Symptoms: It suggests that negative symptoms like poverty of content or alogia might be adaptive responses to excessive noise, rather than pure deficits.
  • Modeling Potential: It lays out how information theory metrics can be used to develop precise, testable computational models of schizophrenia-related dysfunctions.

Limitations and Future Work

The authors acknowledge that this is an early-stage project.

  • Lack of Formal Modeling: While the paper proposes information-theoretic models, no formal computational models based on these concepts for schizophrenia have yet been developed.
  • Explaining Positive Symptoms: The current framework has been primarily applied to perceptual and cognitive deficits. Its utility for explaining positive symptoms like hallucinations and delusions remains to be explored.
  • Negative Symptoms: While a potential explanation for some negative symptoms is offered, their direct modeling using these information theory concepts is an open area.
  • Neurobiological Implementation: While connections to neurobiology are made, the precise mechanisms by which pyramidal neurons implement goal functions like coherent infomax or predictive coding require further clarification.
  • Comparison to Other Models: The authors suggest that future work should compare information-theoretic models with other computational approaches to schizophrenia to determine their relative strengths and weaknesses.

Future work should focus on developing and testing these information-theoretic models empirically, exploring their application to a wider range of symptoms, and further elucidating their neurobiological underpinnings. The authors emphasize the potential for these approaches to guide future theory, experiments, and treatment development.

Summary updated

Abstract

Information theory provides a formal framework within which information processing and its disorders can be described. However, information theory has rarely been applied to modeling aspects of the cognitive neuroscience of schizophrenia. The goal of this article is to highlight the benefits of an approach based on information theory, including its recent extensions, for understanding several disrupted neural goal functions as well as related cognitive and symptomatic phenomena in schizophrenia. We begin by demonstrating that foundational concepts from information theory-such as Shannon information, entropy, data compression, block coding, and strategies to increase the signal-to-noise ratio-can be used to provide novel understandings of cognitive impairments in schizophrenia and metrics to evaluate their integrity. We then describe more recent developments in information theory, including the concepts of infomax, coherent infomax, and coding with synergy, to demonstrate how these can be used to develop computational models of schizophrenia-related failures in the tuning of sensory neurons, gain control, perceptual organization, thought organization, selective attention, context processing, predictive coding, and cognitive control. Throughout, we demonstrate how disordered mechanisms may explain both perceptual/cognitive changes and symptom emergence in schizophrenia. Finally, we demonstrate that there is consistency between some information-theoretic concepts and recent discoveries in neurobiology, especially involving the existence of distinct sites for the accumulation of driving input and contextual information prior to their interaction. This convergence can be used to guide future theory, experiment, and treatment development.
BibTeX
@inproceedings{silverstein2017implications,
  author = {Steven M. Silverstein and Michael Wibral and William A. Phillips},
  title = {{Implications of Information Theory for Computational Modeling of Schizophrenia}},
  booktitle = {Computational},
  year = {2017},
  doi = {10.1162/cpsy\_a\_00004},
}
Figure