Cognitive computational neuroscience (1807.11819v1)

Abstract: To learn how cognition is implemented in the brain, we must build computational models that can perform cognitive tasks, and test such models with brain and behavioral experiments. Cognitive science has developed computational models of human cognition, decomposing task performance into computational components. However, its algorithms still fall short of human intelligence and are not grounded in neurobiology. Computational neuroscience has investigated how interacting neurons can implement component functions of brain computation. However, it has yet to explain how those components interact to explain human cognition and behavior. Modern technologies enable us to measure and manipulate brain activity in unprecedentedly rich ways in animals and humans. However, experiments will yield theoretical insight only when employed to test brain-computational models. It is time to assemble the pieces of the puzzle of brain computation. Here we review recent work in the intersection of cognitive science, computational neuroscience, and artificial intelligence. Computational models that mimic brain information processing during perceptual, cognitive, and control tasks are beginning to be developed and tested with brain and behavioral data.

• The paper demonstrates that task-performing computational models bridge cognitive science, neuroscience, and AI to elucidate brain mechanisms underlying cognition.
• The authors review recent advances, including deep neural networks and Bayesian frameworks, that approximate neural and cognitive processes.
• The paper highlights future directions such as developing neurobiologically plausible AI models, integrating behavioral and neural data, and promoting collaborative research.

Cognitive Computational Neuroscience: Integrating Cognition and Neural Computation

The paper by Kriegeskorte & Douglas presents a comprehensive review of the burgeoning field of cognitive computational neuroscience (CCN), which aims to bridge the gap between cognitive science, computational neuroscience, and AI. The authors make a cogent argument for the development of task-performing computational models that elucidate the neurobiological basis of cognition. Such models, they argue, are indispensable for advancing our understanding of how the brain implements complex cognitive processes.

Integration of Disciplines

CCN seeks to integrate approaches from three key disciplines:

1. Cognitive Science: Traditionally focused on decomposing cognition into computational components, cognitive science offers top-down insights into human cognition. Notably, Bayesian cognitive models have emerged as powerful frameworks for understanding how prior knowledge is combined with sensory information.
2. Computational Neuroscience: This discipline provides bottom-up insights into how neurons and circuits interact to perform computational functions. Recent advancements have produced models that demonstrate basic computational capabilities, such as sensory coding and decision-making mechanisms.
3. Artificial Intelligence: AI, particularly through machine learning and advances in neural networks, has shown significant promise in modeling intelligent behavior. Deep neural networks that share structural similarities with brain processes have been particularly successful in tackling perceptual and cognitive tasks.

Theoretical and Practical Implications

The intersection of these fields is poised to yield significant theoretical and practical advances. Theoretically, it promises a more comprehensive understanding of brain computations that lead to cognitive functions. Practically, insights from CCN could inform the development of AI systems that emulate human-like intelligence and adaptivity.

Recent Developments and Challenges

The authors highlight several recent developments:

• Task-Performing Computational Models: These models are advancing, supported by interdisciplinary research efforts. However, fully neurobiologically plausible models that can be mapped onto brain activity are still in nascent stages.
• Large-Scale Neural Network Models: Studies have begun to test deep neural networks against brain activity data, revealing similarities in representational patterns between these models and primate visual cortex regions. This approach underscores the potential of neural networks as approximations of cortical processes.
• Bayesian Cognitive Models: These models offer a normative framework for understanding cognitive processes, especially in scenarios requiring probabilistic inference. Although promising, current models often struggle with the computational demands of real-world scenarios.

A notable challenge lies in balancing cognitive fidelity with biological plausibility. Models that excel in capturing cognitive processes often do not map well onto neural architectures, and vice versa. Bridging this gap necessitates models that incorporate both behavioral and neurological data.

Future Directions

The paper suggests several directions for future research:

• Neurobiologically Plausible AI Models: Continued collaboration between AI and neuroscience could result in models that not only perform tasks but also align with neural data.
• Integration Across Levels of Analysis: A key goal is to integrate insights across Marr's levels of analysis computational theory, algorithm and representation, and neural implementation.
• Collaborative Research and Open Science: The authors advocate for a collaborative research culture with shared datasets, models, and testing frameworks across disciplines to foster progress in CCN.

In summary, cognitive computational neuroscience holds promise as a field that can provide insights into the algorithms and representations of cognition and brain processes. By marrying the strengths of cognitive science, computational neuroscience, and AI, CCN aims to unravel the complex mechanisms underlying intelligent behavior.