Reconstructing Computational Dynamics from Neural Measurements with Recurrent Neural Networks

Durstewitz, Daniel; Koppe, Georgia; Thurm, Max Ingo

doi:10.1101/2022.10.31.514408

Cited by 10 publications

(17 citation statements)

References 183 publications

(272 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We used trained recurrent neural networks [19, 21, 23]. RNNs can be trained to reproduce both neural activity and behavior [16, 22, 29, 32, 35].…”

Section: Discussionmentioning

confidence: 99%

“…Specifically, we seek to characterize dynamical systems that could underlie working memory relying on phase coding with neural oscillations. We do so by assuming that cognitive functions can be described by a low-dimensional dynamical system, implemented through populations of neurons (computation through dynamics) [18][19][20][21][22][23], in line with empirical observations [24,25].…”

Section: Introductionmentioning

confidence: 99%

“…RNNs are universal dynamical systems, in the sense that they can approximate finite time trajectories of any dynamical system [26,27]. We are thus able to use these networks as hypothesis generators -first training them on a cognitive task, and then reverse engineering the resulting networks [18,22,[28][29][30][31][32][33][34][35].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Trained recurrent neural networks develop phase-locked limit cycles in a working memory task

Pals

Macke

Barak

2023

Preprint

View full text Add to dashboard Cite

Neural oscillations are ubiquitously observed in many brain areas. One proposed functional role of these oscillations is that they serve as an internal clock, or 'frame of reference'. Information can be encoded by the timing of neural activity relative to the phase of such oscillations. In line with this hypothesis, there have been multiple empirical observations of such phase codes in the brain. Here we ask: What kind of neural dynamics support phase coding of information with neural oscillations? We tackled this question by analyzing recurrent neural networks (RNNs) that were trained on a working memory task. The networks were given access to an external reference oscillation and tasked to produce an oscillation, such that the phase difference between the reference and output oscillation maintains the identity of transient stimuli. We found that networks converged to stable oscillatory dynamics. Reverse engineering these networks revealed that each phase-coded memory corresponds to a separate limit cycle attractor. We characterized how the stability of the attractor dynamics depends on both reference oscillation amplitude and frequency, properties that can be experimentally observed. To understand the connectivity structures that underlie these dynamics, we showed that trained networks can be described as two phase-coupled oscillators. Using this insight, we condensed our trained networks to a reduced model consisting of two functional modules: One that generates an oscillation and one that implements a coupling function between the internal oscillation and external reference. In summary, by reverse engineering the dynamics and connectivity of trained RNNs, we propose a mechanism by which neural networks can harness reference oscillations for working memory. Specifically, we propose that a phase-coding network generates autonomous oscillations which it couples to an external reference oscillation in a multi-stable fashion.

show abstract

“…We used trained recurrent neural networks [19, 21, 23]. RNNs can be trained to reproduce both neural activity and behavior [16, 22, 29, 32, 35].…”

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Trained recurrent neural networks develop phase-locked limit cycles in a working memory task

Pals

Macke

Barak

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…In addition, this work also selects common models such as SVM, RNN, CNN (2D CNN), and LSTM for comparative analysis [40,41] TP represents the number of samples correctly predicted as the positive class, TN is the number of samples correctly predicted as the negative class, FP is the number of samples actually belonging to the negative class but wrongly predicted as the positive class, and FN is the number of samples actually belonging to the positive class but wrongly predicted as the negative class.…”

Section: Convolutional Neural Networkmentioning

confidence: 99%

Emotion Recognition and Management in the Tourism Industry During Emergency Events Using Improved Convolutional Neural Network

Qi,

Han

2024

IEEE Access

View full text Add to dashboard Cite

The purpose is to optimize emotion recognition capabilities in the tourism industry during emergency events and enhance management efficiency. Hence, based on the principles of emergency events and emotion recognition, this work has outlined the fundamental emotion recognition process. Additionally, an emotion recognition model is constructed based on a Convolutional Neural Network (CNN), and the process of extracting emotions using a three-dimensional (3D) CNN is proposed. Finally, an attention mechanism is employed to optimize the 3D CNN model, and a comparative analysis of accuracy and precision is conducted using the Bimodal Face and Body Gesture Database (FABO). The research findings reveal that the optimized 3D CNN model exhibits lower error rates in recognizing anxiety emotions, with only 3 samples going unrecognized. Its overall recognition performance is superior to other models. There are variations in the recognition accuracy among different models, but in general, the optimized 3D CNN model performs relatively well across various datasets, achieving recognition accuracies of 83.92%, 82.33%, and 88.81%. Compared to other models, the optimized 3D CNN model demonstrates higher precision in recognizing different emotions, particularly excelling in identifying anger, disgust, and happiness, with precision rates of 97%, 91%, and 94%, respectively. This work has improved the accuracy and efficiency of emotion recognition, providing more intelligent and effective support for emergency event management in the tourism industry.

show abstract

“…We used these gradients to build a 5-d coordinate system that allows us to organize whole-brain maps in a ‘common space’, in which the relative locations within this space provide information regarding the balance of different macroscale systems in a particular context (see also 14 – 16 ). This analytic approach is focused on how different large-scale systems interact together and so provides a biologically relevant macroscale perspective on brain states 17 that complements perspectives that focus on parcels 18 and large-scale networks 19 (see Konu et al 11 for a region-based analysis of the data used in the current study).…”

Section: Introductionmentioning

confidence: 99%

Experience sampling reveals the role that covert goal states play in task-relevant behavior

Mckeown,

Strawson,

Zhang

et al. 2023

Sci Rep

View full text Add to dashboard Cite

Cognitive neuroscience has gained insight into covert states using experience sampling. Traditionally, this approach has focused on off-task states. However, task-relevant states are also maintained via covert processes. Our study examined whether experience sampling can also provide insights into covert goal-relevant states that support task performance. To address this question, we developed a neural state space, using dimensions of brain function variation, that allows neural correlates of overt and covert states to be examined in a common analytic space. We use this to describe brain activity during task performance, its relation to covert states identified via experience sampling, and links between individual variation in overt and covert states and task performance. Our study established deliberate task focus was linked to faster target detection, and brain states underlying this experience—and target detection—were associated with activity patterns emphasizing the fronto-parietal network. In contrast, brain states underlying off-task experiences—and vigilance periods—were linked to activity patterns emphasizing the default mode network. Our study shows experience sampling can not only describe covert states that are unrelated to the task at hand, but can also be used to highlight the role fronto-parietal regions play in the maintenance of covert task-relevant states.

show abstract

Reconstructing Computational Dynamics from Neural Measurements with Recurrent Neural Networks

Cited by 10 publications

References 183 publications

Trained recurrent neural networks develop phase-locked limit cycles in a working memory task

Trained recurrent neural networks develop phase-locked limit cycles in a working memory task

Emotion Recognition and Management in the Tourism Industry During Emergency Events Using Improved Convolutional Neural Network

Experience sampling reveals the role that covert goal states play in task-relevant behavior

Contact Info

Product

Resources

About