Predictive coding explains asymmetric connectivity in the brain: A neural network study
Clinical Snapshot
PICO Framework
| P — Population | Computational models (convolutional neural networks augmented with predictive feedback), used as proxies for biological visual processing systems; no human or animal subjects directly studied |
| I — Intervention | Asymmetric predictive coding feedback connections of varying ranges (short-, medium-, long-, and longer-range) originating at the same source layer or terminating at the same target layer, incorporated into CNN models |
| C — Comparator | Symmetric predictive feedback connections (baseline augmented model) and feedforward-only CNN models (no feedback) |
| O — Outcomes | Grasp detection performance (accuracy/robustness) on noisy visual stimuli; characterisation of which feedback configuration optimally enhances model robustness under adverse visual conditions |
Bottom Line
This computational neuroscience study from the University of Toronto uses convolutional neural networks augmented with predictive coding feedback to investigate why the brain's cortical feedback connections are asymmetric rather than symmetric, as classical predictive coding theory assumes. The authors find that medium-range asymmetric feedback — particularly when originating from layers close to the input — most effectively improves model robustness to noisy visual stimuli in a grasp detection task. This is an intellectually coherent and methodologically sound basic science contribution that advances mechanistic understanding of cortical hierarchy and predictive processing. However, the study is entirely computational, with no human or animal subjects, no clinical outcomes, and no quantitative effect sizes or confidence intervals reported in the abstract. It cannot be translated into clinical recommendations. Clinicians and neuroscientists should regard this as hypothesis-generating work that may, over time, inform the design of neuroprosthetic devices or visual rehabilitation systems. Independent replication, validation against empirical neurophysiological data, and extension beyond the grasp detection domain are necessary before broader conclusions can be drawn.
Key Findings
P Value: Not reported in abstract
Effect Size: Not reported in abstract; specific performance metrics (e.g., accuracy percentages, detection rates) are not provided
Primary Outcome: Medium-range asymmetric predictive feedback connections optimally enhanced grasp detection performance under noisy visual conditions compared to short-, long-, and longer-range configurations
Nnt Or Sensitivity: Not applicable (computational modelling study); no NNT, sensitivity, specificity, or hazard ratio reported. Performance improvement is described qualitatively as 'optimal' for medium-range feedback originating proximal to the input layer
Confidence Interval: Not reported
Clinical Application
No immediate clinical implementation is feasible or appropriate. The study provides mechanistic insights that may, in the long term, inform the design of neuroprosthetic systems, visual rehabilitation technologies, or neuromodulation strategies targeting feedback pathways in conditions such as cortical visual impairment or stroke-related visuomotor deficits This study has no direct relevance to current Australian PBS listings, TGA-approved therapeutics, or RACGP clinical guidelines. In the Australian context, it may be of interest to researchers at institutions such as the Australian Brain Alliance, Monash University's Turner Institute for Brain and Mental Health, or the University of Melbourne's Grasp and Perception Laboratory. Future translational work could intersect with NHMRC-funded programmes in neurorehabilitation and upper-limb prosthetics. No changes to clinical practice are warranted based on this study alone. Not directly applicable to any clinical patient population at this stage. Findings are relevant to researchers in computational neuroscience, cognitive neuroscience, and potentially to engineers developing brain-computer interfaces or robotic prosthetic limb control systems
Abstract
Seminal frameworks of predictive coding propose a hierarchy of generative modules, each attempting to infer the neural representation of the module one level below; the predictions are carried by top-down feedback projections, while the predictive error is propagated by reciprocal forward pathways. Such symmetric feedback connections support visual processing of noisy stimuli in computational models. However, neurophysiological studies have yielded evidence of asymmetric cortical feedback connections. We investigated the contribution of neural feedback in visual processing for computing grasp parameters, by utilizing convolutional neural network models that had been augmented with predictive feedback and were trained to compute grasp positions for real-world objects. After establishing an ameliorative effect of symmetric feedback on grasp detection performance when evaluated on noisy stimuli, we characterized the performance effects of asymmetric feedback, similar to that observed in the cortex. Specifically, we tested model variants extended with short-, medium-, long- and longer-range feedback connections (i) originating at the same source layer or (ii) terminating at the same target layer. We found that the performance-enhancing effect of predictive coding under adverse conditions was optimal for medium-range asymmetric feedback. Moreover, this effect was most prominent when medium-range feedback originated at a level of representational abstraction that was proximal to the input layer, in contrast to more distal layers. To conclude, our simulations show that introducing biologically realistic asymmetric predictive feedback improves model robustness to noisy visual stimuli in a neural network model optimized for grasp detection.
References
- 1.Khan, R., Zhong, H., Das, S., Cai, J., & Niemeier, M. (2026). Predictive coding explains asymmetric connectivity in the brain: A neural network study. PLoS Computational Biology. https://doi.org/10.1371/journal.pcbi.1014435
Related Research
Current biology : CB
Looking to the brain to improve energy efficiency of AI
21 July 2026
Philosophical transactions of the Royal Society of London. Series B, Biological sciences
The hippocampus as a generative model
11 July 2026
Journal of mathematical biology
Spiking neural models for decision-making tasks with learning.
14 June 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service