Aligned and oblique dynamics in recurrent neural networks

The relation between neural activity and behaviorally relevant variables is at the heart of neuroscience research. When strong, this relation is termed a neural representation. There is increasing evidence, however, for partial dissociations between activity in an area and relevant external variables. While many explanations have been proposed, a theoretical framework for the relationship between external and internal variables is lacking. Here, we utilize recurrent neural networks (RNNs) to explore the question of when and how neural dynamics and the network's output are related from a geometrical point of view. We find that RNNs can operate in two regimes: dynamics can either be aligned with the directions that generate output variables, or oblique to them. We show that the magnitude of the readout weights can serve as a control knob between the regimes. Importantly, these regimes are functionally distinct. Oblique networks are more heterogeneous and suppress noise in their output directions. They are furthermore more robust to perturbations along the output directions. Finally, we show that the two regimes can be dissociated in neural recordings. Altogether, our results open a new perspective for interpreting neural activity by relating network dynamics and their output.

[1]  M. Sahani,et al.  Direct neural perturbations reveal a dynamical mechanism for robust computation , 2022, bioRxiv.

[2]  Jonathan W. Pillow,et al.  A new theoretical framework jointly explains behavioral and neural variability across subjects performing flexible decision-making , 2022, bioRxiv.

[3]  Jacob L. Yates,et al.  Activity in primate visual cortex is minimally driven by spontaneous movements , 2022, bioRxiv.

[4]  Carson C. Chow,et al.  Distributing task-related neural activity across a cortical network through task-independent connections , 2022, bioRxiv.

[5]  Andrew M. Saxe,et al.  Orthogonal representations for robust context-dependent task performance in brains and neural networks , 2022, Neuron.

[6]  O. Barak,et al.  Charting and Navigating the Space of Solutions for Recurrent Neural Networks , 2021, NeurIPS.

[7]  Scott W. Linderman,et al.  Generalized Shape Metrics on Neural Representations , 2021, NeurIPS.

[8]  Eva L. Dyer,et al.  Neural Latents Benchmark '21: Evaluating latent variable models of neural population activity , 2021, NeurIPS Datasets and Benchmarks.

[9]  M. Sahani,et al.  Residual dynamics resolves recurrent contributions to neural computation , 2021, bioRxiv.

[10]  M. Häusser,et al.  All-optical interrogation of neural circuits in behaving mice , 2021, Nature Protocols.

[11]  Arthur Jacot,et al.  Saddle-to-Saddle Dynamics in Deep Linear Networks: Small Initialization Training, Symmetry, and Sparsity , 2021, 2106.15933.

[12]  Francis R. Willett,et al.  High-performance brain-to-text communication via handwriting , 2021, Nature.

[13]  Greg Yang,et al.  Feature Learning in Infinite-Width Neural Networks , 2020, ArXiv.

[14]  O. Barak,et al.  Quality of internal representation shapes learning performance in feedback neural networks , 2020, 2011.06066.

[15]  Mikhail Belkin,et al.  On the linearity of large non-linear models: when and why the tangent kernel is constant , 2020, NeurIPS.

[16]  Krishna V. Shenoy,et al.  Computation Through Neural Population Dynamics. , 2020, Annual review of neuroscience.

[17]  S. Ostojic,et al.  The role of population structure in computations through neural dynamics , 2022, Nature Neuroscience.

[18]  Francesca Mastrogiuseppe,et al.  Shaping Dynamics With Multiple Populations in Low-Rank Recurrent Networks , 2020, Neural Computation.

[19]  Surya Ganguli,et al.  Predictive coding in balanced neural networks with noise, chaos and delays , 2020, NeurIPS.

[20]  Francesca Mastrogiuseppe,et al.  The interplay between randomness and structure during learning in RNNs , 2020, NeurIPS.

[21]  Byron M. Yu,et al.  Stabilization of a brain–computer interface via the alignment of low-dimensional spaces of neural activity , 2020, Nature Biomedical Engineering.

[22]  俊一 甘利 5分で分かる!? 有名論文ナナメ読み:Jacot, Arthor, Gabriel, Franck and Hongler, Clement : Neural Tangent Kernel : Convergence and Generalization in Neural Networks , 2020 .

[23]  Guillaume Hennequin,et al.  Optimal anticipatory control as a theory of motor preparation: A thalamo-cortical circuit model , 2020, Neuron.

[24]  Sandro Romani,et al.  Attractor dynamics gate cortical information flow during decision-making , 2019, Nature Neuroscience.

[25]  Lee E. Miller,et al.  Long-term stability of cortical population dynamics underlying consistent behavior , 2019, Nature Neuroscience.

[26]  O. Barak,et al.  Dynamics of random recurrent networks with correlated low-rank structure , 2019, 1909.04358.

[27]  Matthew T. Kaufman,et al.  Single-trial neural dynamics are dominated by richly varied movements , 2019, Nature Neuroscience.

[28]  Surya Ganguli,et al.  Universality and individuality in neural dynamics across large populations of recurrent networks , 2019, NeurIPS.

[29]  Byron M. Yu,et al.  New neural activity patterns emerge with long-term learning , 2019, Proceedings of the National Academy of Sciences.

[30]  Abigail A. Russo,et al.  Neural Trajectories in the Supplementary Motor Area and Motor Cortex Exhibit Distinct Geometries, Compatible with Different Classes of Computation , 2019, Neuron.

[31]  Shreya Saxena,et al.  Towards the neural population doctrine , 2019, Current Opinion in Neurobiology.

[32]  Selmaan N. Chettih,et al.  Single-neuron perturbations reveal feature-specific competition in V1 , 2019, Nature.

[33]  Francis Bach,et al.  On Lazy Training in Differentiable Programming , 2018, NeurIPS.

[34]  Christian Ethier,et al.  Cortical population activity within a preserved neural manifold underlies multiple motor behaviors , 2018, Nature Communications.

[35]  Byron M. Yu,et al.  Constraints on neural redundancy , 2018, eLife.

[36]  Nicholas A. Steinmetz,et al.  High-dimensional geometry of population responses in visual cortex , 2018, bioRxiv.

[37]  Nicholas A. Steinmetz,et al.  Spontaneous behaviors drive multidimensional, brainwide activity , 2019, Science.

[38]  Andrea Montanari,et al.  A mean field view of the landscape of two-layer neural networks , 2018, Proceedings of the National Academy of Sciences.

[39]  Abigail A. Russo,et al.  Motor Cortex Embeds Muscle-like Commands in an Untangled Population Response , 2018, Neuron.

[40]  Byron M. Yu,et al.  Learning by neural reassociation , 2018, Nature Neuroscience.

[41]  Joseph G. Makin,et al.  Superior arm-movement decoding from cortex with a new, unsupervised-learning algorithm , 2018, Journal of neural engineering.

[42]  Francesca Mastrogiuseppe,et al.  Linking Connectivity, Dynamics, and Computations in Low-Rank Recurrent Neural Networks , 2017, Neuron.

[43]  Surya Ganguli,et al.  A theory of multineuronal dimensionality, dynamics and measurement , 2017, bioRxiv.

[44]  Luca Antiga,et al.  Automatic differentiation in PyTorch , 2017 .

[45]  Omri Barak,et al.  Recurrent neural networks as versatile tools of neuroscience research , 2017, Current Opinion in Neurobiology.

[46]  Lee E. Miller,et al.  Neural Manifolds for the Control of Movement , 2017, Neuron.

[47]  Victor M. Preciado,et al.  Controllability Gramian spectra of random networks , 2016, 2016 American Control Conference (ACC).

[48]  Moritz Helias,et al.  Optimal Sequence Memory in Driven Random Networks , 2016, Physical Review X.

[49]  G. Cheng,et al.  Adjoint methods of sensitivity analysis for Lyapunov equation , 2016 .

[50]  Alexander Rivkind,et al.  Local Dynamics in Trained Recurrent Neural Networks. , 2015, Physical review letters.

[51]  Matthew T. Kaufman,et al.  A neural network that finds a naturalistic solution for the production of muscle activity , 2015, Nature Neuroscience.

[52]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[53]  Byron M. Yu,et al.  Neural constraints on learning , 2014, Nature.

[54]  W. Gerstner,et al.  Optimal Control of Transient Dynamics in Balanced Networks Supports Generation of Complex Movements , 2014, Neuron.

[55]  David Sussillo,et al.  Neural circuits as computational dynamical systems , 2014, Current Opinion in Neurobiology.

[56]  Matthew T. Kaufman,et al.  Cortical activity in the null space: permitting preparation without movement , 2014, Nature Neuroscience.

[57]  W. Newsome,et al.  Context-dependent computation by recurrent dynamics in prefrontal cortex , 2013, Nature.

[58]  Xiao-Jing Wang,et al.  The importance of mixed selectivity in complex cognitive tasks , 2013, Nature.

[59]  Matthew T. Kaufman,et al.  Neural population dynamics during reaching , 2012, Nature.

[60]  Matthew T. Kaufman,et al.  Cortical Preparatory Activity: Representation of Movement or First Cog in a Dynamical Machine? , 2010, Neuron.

[61]  L. F. Abbott,et al.  Generating Coherent Patterns of Activity from Chaotic Neural Networks , 2009, Neuron.

[62]  M. Scanziani,et al.  Instantaneous Modulation of Gamma Oscillation Frequency by Balancing Excitation with Inhibition , 2009, Neuron.

[63]  W. Maass,et al.  State-dependent computations: spatiotemporal processing in cortical networks , 2009, Nature Reviews Neuroscience.

[64]  Michael Okun,et al.  Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities , 2008, Nature Neuroscience.

[65]  E. Bizzi,et al.  Motor Learning with Unstable Neural Representations , 2007, Neuron.

[66]  T. Hafting,et al.  Microstructure of a spatial map in the entorhinal cortex , 2005, Nature.

[67]  R. Romo,et al.  Neuronal correlates of parametric working memory in the prefrontal cortex , 1999, Nature.

[68]  A. Grinvald,et al.  Dynamics of Ongoing Activity: Explanation of the Large Variability in Evoked Cortical Responses , 1996, Science.

[69]  Sommers,et al.  Chaos in random neural networks. , 1988, Physical review letters.

[70]  J. O'Keefe,et al.  The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. , 1971, Brain research.

[71]  T. Faniran Numerical Solution of Stochastic Differential Equations , 2015 .

[72]  D. Hubel,et al.  Receptive fields, binocular interaction and functional architecture in the cat's visual cortex , 1962, The Journal of physiology.