Imprecise neural computations as source of human adaptive behavior in volatile environments

Findling, Charles; Chopin, Nicolás; Koechlin, Etienne

doi:10.1101/799239

Cited by 7 publications

(12 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, during testing, our artificial agents became observers bound to integrate the presented cues in the weather prediction task, whereas they remained agents seeking rewards through interaction with their environment in the bandit task 32,38 . Despite these fundamental differences, we found that computation noise promotes very similar functions during reasoning and meta-learning -two cognitive abilities associated with substantial computation noise in humans [18][19][20] .…”

Section: Discussionmentioning

confidence: 68%

“…The structure of the computation noise we implemented in artificial neural networks was directly inspired by these recent observations in humans [18][19][20] . But instead of relying on an 'algorithmic' description of underlying computations as in earlier work (Bayesian inference in the weather prediction task, reinforcement learning in the bandit task), we adopted a neural network description which offers several advantages 29,30 .…”

Section: Discussionmentioning

confidence: 99%

“…As before, we varied the level of computation noise (fixed across training and testing) in the noisy RNNs, and the amount of explicit action entropy regularization in the exact RNNs (see Methods). Based on recent observations of such noise in humans 19,20 , computation noise during meta-learning was implemented by corrupting the updates of each unit in the network with i.i.d. normally distributed noise whose standard deviation scales with the quantity of update itself.…”

Section: Lossless Recovery From Misleading Outcomes In Noisy Meta-leamentioning

confidence: 99%

“…Beyond input noise, biological neural networks process and respond to input with a large internally generated variability 15 that is absent from artificial neural networks whose units exhibit deterministic input-state-output relations after training. This 'computation noise' has a wideranging impact on human cognition and behavior 16 -from fluctuations in the perception of weak sensory stimuli 17 to 'exploratory' behavior during reward-guided decision-making [18][19][20] . Existing research has typically considered this internal noise as a hard constraint on neural information processing systems, that the brain has evolved to cope with using efficient coding strategies [21][22][23][24] .…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Computation noise promotes cognitive resilience to adverse conditions during decision-making

Findling

Wyart

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Random noise in information processing systems is widely seen as detrimental to function. But despite the large trial-to-trial variability of neural activity and behavior, humans and other animals show a remarkable adaptability to unexpected adverse events occurring during task execution. This cognitive ability, described as constitutive of general intelligence, is missing from current artificial intelligence (AI) systems which feature exact (noise-free) computations. Here we show that implementing computation noise in recurrent neural networks boosts their cognitive resilience to a variety of adverse conditions entirely unseen during training, in a way that resembles human and animal cognition. In contrast to artificial agents with exact computations, noisy agents exhibit hallmarks of Bayesian inference acquired in a 'zero-shot' fashion -without prior experience with conditions that require these computations for maximizing rewards. We further demonstrate that these cognitive benefits result from free-standing regularization of activity patterns in noisy neural networks. Together, these findings suggest that intelligence may ride on computation noise to promote near-optimal decisionmaking in adverse conditions without any engineered cognitive sophistication.

show abstract

Section: Discussionmentioning

confidence: 68%

Section: Discussionmentioning

confidence: 99%

Section: Lossless Recovery From Misleading Outcomes In Noisy Meta-leamentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Computation noise promotes cognitive resilience to adverse conditions during decision-making

Findling

Wyart

2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…For example, in the context of fMRI investigations of value-based decisions, one may search for brain regions whose activity eventually perturbs the computation and/or comparison of options’ values. This would extend the portfolio of recent empirical studies of neural noise perturbations to learning-relevant computations (Drugowitsch et al, 2016; Findling et al, 2019; Wyart and Koechlin, 2016). Reciprocally, using some variant of mediation analysis (Brochard and Daunizeau, 2020; Lindquist, 2012; MacKinnon et al, 2007), one may extract neuroimaging estimates of neural noise that can inform DDM-based behavioral data analysis.…”

Section: Discussionmentioning

confidence: 92%

An overcomplete approach to fitting drift-diffusion decision models to trial-by-trial data

Feltgen¹,

Daunizeau

2020

Preprint

View full text Add to dashboard Cite

Drift-diffusion models or DDMs are becoming a standard in the field of computational neuroscience. They extend models from signal detection theory by proposing a simple mechanistic explanation for the observed relationship between decision outcomes and reaction times (RT). In brief, they assume that decisions are triggered once the accumulated evidence in favor of a particular alternative option has reached a predefined threshold. Fitting a DDM to empirical data then allows one to interpret observed group or condition differences in terms of a change in the underlying model parameters. However, current approaches do not provide reliable parameter estimates when, e.g., evidence strength is varied over trials. In this note, we propose a fast and efficient approach that is based on fitting a self-consistency equation that the DDM fulfills. Using numerical simulations, we show that this approach enables one to extract relevant information from trial-by-trial variations of RT data that would typically be buried in the empirical distribution. Finally, we demonstrate the added-value of the approach, when applied to a recent value-based decision making experiment.

show abstract

Volatility of subliminal haptic feedback alters the feeling of control in schizophrenia.

Foerster¹,

Weibel²,

Poncelet³

et al. 2021

Journal of Abnormal Psychology

View full text Add to dashboard Cite

It has been proposed that agency disorders found in schizophrenia rely on aberrant processing of prediction error. Overreactivity to nonpertinent prediction errors may lead to the attribution of one's own actions to an external source. When applied to perception, this could explain hallucinations. However, experiments in motor control or perception have mainly suggested deficient prediction errors. Using a novel approach based on the manipulation of temporal delays, 23 patients with schizophrenia, 18 patients with bipolar disorder, and 22 healthy participants performed a pointing task with a haptic device that provided haptic feedback without or with delays, which were processed consciously (65 ms) or unconsciously (15 ms). The processing of prediction errors was measured via the adaptation of the hand trajectory, that is, the deceleration in anticipation of the surface, and its modulation as a function of recent history (stable or unstable sensory feedback). Agency was evaluated by measuring the participants' feeling of controlling the device. Only patients with schizophrenia reported a decrease in the feeling of control following subliminally delayed haptic feedback and adapted deceleration durations following subliminally delayed haptic feedback. This effect was correlated with positive symptoms. The overreactivity to subliminal delays was present only when delays occurred repeatedly in an unpredictable way, that is, with a volatile distribution. The results suggest that small temporal uncertainties that should be held as negligible, trigger an aberrant overreactivity which could account for hallucinations and alterations of the patients' conscious feeling of control. General Scientific SummaryInsignificantly small temporal uncertainties in the perceptual and motor domains are typically ignored. This study reveals that patients with schizophrenia, but not bipolar disorders, overreact to delays of a few milliseconds in the haptic feedback. They inadequately adapt their movements to delays and their feeling of control decreases. These effects correlate with positive symptoms, suggesting that negligible prediction errors may become salient in an aberrant way and may thus disturb agency.

show abstract

Imprecise neural computations as source of human adaptive behavior in volatile environments

Cited by 7 publications

References 36 publications

Computation noise promotes cognitive resilience to adverse conditions during decision-making

Computation noise promotes cognitive resilience to adverse conditions during decision-making

An overcomplete approach to fitting drift-diffusion decision models to trial-by-trial data

Volatility of subliminal haptic feedback alters the feeling of control in schizophrenia.

Contact Info

Product

Resources

About