Recurrent neural networks that learn multi-step visual routines with reinforcement learning.

Many cognitive problems can be decomposed into series of subproblems that are solved sequentially by the brain. When subproblems are solved, relevant intermediate results need to be stored by neurons and propagated to the next subproblem, until the overarching goal has been completed. We will here consider visual tasks, which can be decomposed into sequences of elemental visual operations. Experimental evidence suggests that intermediate results of the elemental operations are stored in working memory as an enhancement of neural activity in the visual cortex. The focus of enhanced activity is then available for subsequent operations to act upon. The main question at stake is how the elemental operations and their sequencing can emerge in neural networks that are trained with only rewards, in a reinforcement learning setting. We here propose a new recurrent neural network architecture that can learn composite visual tasks that require the application of successive elemental operations. Specifically, we selected three tasks for which electrophysiological recordings of monkeys' visual cortex are available. To train the networks, we used RELEARNN, a biologically plausible four-factor Hebbian learning rule, which is local both in time and space. We report that networks learn elemental operations, such as contour grouping and visual search, and execute sequences of operations, solely based on the characteristics of the visual stimuli and the reward structure of a task. After training was completed, the activity of the units of the neural network elicited by behaviorally relevant image items was stronger than that elicited by irrelevant ones, just as has been observed in the visual cortex of monkeys solving the same tasks. Relevant information that needed to be exchanged between subroutines was maintained as a focus of enhanced activity and passed on to the subsequent subroutines. Our results demonstrate how a biologically plausible learning rule can train a recurrent neural network on multistep visual tasks.

Mechanisms of human dynamic object recognition revealed by sequential deep neural networks.

Humans can quickly recognize objects in a dynamically changing world. This ability is showcased by the fact that observers succeed at recognizing objects in rapidly changing image sequences, at up to 13 ms/image. To date, the mechanisms that govern dynamic object recognition remain poorly understood. Here, we developed deep learning models for dynamic recognition and compared different computational mechanisms, contrasting feedforward and recurrent, single-image and sequential processing as well as different forms of adaptation. We found that only models that integrate images sequentially via lateral recurrence mirrored human performance (N = 36) and were predictive of trial-by-trial responses across image durations (13-80 ms/image). Importantly, models with sequential lateral-recurrent integration also captured how human performance changes as a function of image presentation durations, with models processing images for a few time steps capturing human object recognition at shorter presentation durations and models processing images for more time steps capturing human object recognition at longer presentation durations. Furthermore, augmenting such a recurrent model with adaptation markedly improved dynamic recognition performance and accelerated its representational dynamics, thereby predicting human trial-by-trial responses using fewer processing resources. Together, these findings provide new insights into the mechanisms rendering object recognition so fast and effective in a dynamic visual world.

Arousal state affects perceptual decision-making by modulating hierarchical sensory processing in a large-scale visual system model.

Arousal levels strongly affect task performance. Yet, what arousal level is optimal for a task depends on its difficulty. Easy task performance peaks at higher arousal levels, whereas performance on difficult tasks displays an inverted U-shape relationship with arousal, peaking at medium arousal levels, an observation first made by Yerkes and Dodson in 1908. It is commonly proposed that the noradrenergic locus coeruleus system regulates these effects on performance through a widespread release of noradrenaline resulting in changes of cortical gain. This account, however, does not explain why performance decays with high arousal levels only in difficult, but not in simple tasks. Here, we present a mechanistic model that revisits the Yerkes-Dodson effect from a sensory perspective: a deep convolutional neural network augmented with a global gain mechanism reproduced the same interaction between arousal state and task difficulty in its performance. Investigating this model revealed that global gain states differentially modulated sensory information encoding across the processing hierarchy, which explained their differential effects on performance on simple versus difficult tasks. These findings offer a novel hierarchical sensory processing account of how, and why, arousal state affects task performance.

A Probabilistic Digital Twin for Leak Localization in Water Distribution Networks Using Generative Deep Learning.

Localizing leakages in large water distribution systems is an important and ever-present problem. Due to the complexity originating from water pipeline networks, too few sensors, and noisy measurements, this is a highly challenging problem to solve. In this work, we present a methodology based on generative deep learning and Bayesian inference for leak localization with uncertainty quantification. A generative model, utilizing deep neural networks, serves as a probabilistic surrogate model that replaces the full equations, while at the same time also incorporating the uncertainty inherent in such models. By embedding this surrogate model into a Bayesian inference scheme, leaks are located by combining sensor observations with a model output approximating the true posterior distribution for possible leak locations. We show that our methodology enables producing fast, accurate, and trustworthy results. It showed a convincing performance on three problems with increasing complexity. For a simple test case, the Hanoi network, the average topological distance (ATD) between the predicted and true leak location ranged from 0.3 to 3 with a varying number of sensors and level of measurement noise. For two more complex test cases, the ATD ranged from 0.75 to 4 and from 1.5 to 10, respectively. Furthermore, accuracies upwards of 83%, 72%, and 42% were achieved for the three test cases, respectively. The computation times ranged from 0.1 to 13 s, depending on the size of the neural network employed. This work serves as an example of a digital twin for a sophisticated application of advanced mathematical and deep learning techniques in the area of leak detection.

Leveraging Spiking Deep Neural Networks to Understand the Neural Mechanisms Underlying Selective Attention.

Spatial attention enhances sensory processing of goal-relevant information and improves perceptual sensitivity. Yet, the specific neural mechanisms underlying the effects of spatial attention on performance are still contested. Here, we examine different attention mechanisms in spiking deep convolutional neural networks. We directly contrast effects of precision (internal noise suppression) and two different gain modulation mechanisms on performance on a visual search task with complex real-world images. Unlike standard artificial neurons, biological neurons have saturating activation functions, permitting implementation of attentional gain as gain on a neuron's input or on its outgoing connection. We show that modulating the connection is most effective in selectively enhancing information processing by redistributing spiking activity and by introducing additional task-relevant information, as shown by representational similarity analyses. Precision only produced minor attentional effects in performance. Our results, which mirror empirical findings, show that it is possible to adjudicate between attention mechanisms using more biologically realistic models and natural stimuli.

Flexible Working Memory Through Selective Gating and Attentional Tagging.

Working memory is essential: it serves to guide intelligent behavior of humans and nonhuman primates when task-relevant stimuli are no longer present to the senses. Moreover, complex tasks often require that multiple working memory representations can be flexibly and independently maintained, prioritized, and updated according to changing task demands. Thus far, neural network models of working memory have been unable to offer an integrative account of how such control mechanisms can be acquired in a biologically plausible manner. Here, we present WorkMATe, a neural network architecture that models cognitive control over working memory content and learns the appropriate control operations needed to solve complex working memory tasks. Key components of the model include a gated memory circuit that is controlled by internal actions, encoding sensory information through untrained connections, and a neural circuit that matches sensory inputs to memory content. The network is trained by means of a biologically plausible reinforcement learning rule that relies on attentional feedback and reward prediction errors to guide synaptic updates. We demonstrate that the model successfully acquires policies to solve classical working memory tasks, such as delayed recognition and delayed pro-saccade/anti-saccade tasks. In addition, the model solves much more complex tasks, including the hierarchical 12-AX task or the ABAB ordered recognition task, both of which demand an agent to independently store and updated multiple items separately in memory. Furthermore, the control strategies that the model acquires for these tasks subsequently generalize to new task contexts with novel stimuli, thus bringing symbolic production rule qualities to a neural network architecture. As such, WorkMATe provides a new solution for the neural implementation of flexible memory control.

Depth in convolutional neural networks solves scene segmentation.

Feed-forward deep convolutional neural networks (DCNNs) are, under specific conditions, matching and even surpassing human performance in object recognition in natural scenes. This performance suggests that the analysis of a loose collection of image features could support the recognition of natural object categories, without dedicated systems to solve specific visual subtasks. Research in humans however suggests that while feedforward activity may suffice for sparse scenes with isolated objects, additional visual operations ('routines') that aid the recognition process (e.g. segmentation or grouping) are needed for more complex scenes. Linking human visual processing to performance of DCNNs with increasing depth, we here explored if, how, and when object information is differentiated from the backgrounds they appear on. To this end, we controlled the information in both objects and backgrounds, as well as the relationship between them by adding noise, manipulating background congruence and systematically occluding parts of the image. Results indicate that with an increase in network depth, there is an increase in the distinction between object- and background information. For more shallow networks, results indicated a benefit of training on segmented objects. Overall, these results indicate that, de facto, scene segmentation can be performed by a network of sufficient depth. We conclude that the human brain could perform scene segmentation in the context of object identification without an explicit mechanism, by selecting or "binding" features that belong to the object and ignoring other features, in a manner similar to a very deep convolutional neural network.

Using the structure of genome data in the design of deep neural networks for predicting amyotrophic lateral sclerosis from genotype.

MOTIVATION: Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease caused by aberrations in the genome. While several disease-causing variants have been identified, a major part of heritability remains unexplained. ALS is believed to have a complex genetic basis where non-additive combinations of variants constitute disease, which cannot be picked up using the linear models employed in classical genotype-phenotype association studies. Deep learning on the other hand is highly promising for identifying such complex relations. We therefore developed a deep-learning based approach for the classification of ALS patients versus healthy individuals from the Dutch cohort of the Project MinE dataset. Based on recent insight that regulatory regions harbor the majority of disease-associated variants, we employ a two-step approach: first promoter regions that are likely associated to ALS are identified, and second individuals are classified based on their genotype in the selected genomic regions. Both steps employ a deep convolutional neural network. The network architecture accounts for the structure of genome data by applying convolution only to parts of the data where this makes sense from a genomics perspective. RESULTS: Our approach identifies potentially ALS-associated promoter regions, and generally outperforms other classification methods. Test results support the hypothesis that non-additive combinations of variants contribute to ALS. Architectures and protocols developed are tailored toward processing population-scale, whole-genome data. We consider this a relevant first step toward deep learning assisted genotype-phenotype association in whole genome-sized data. AVAILABILITY AND IMPLEMENTATION: Our code will be available on Github, together with a synthetic dataset (https://github.com/byin-cwi/ALS-Deeplearning). The data used in this study is available to bona-fide researchers upon request. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

How attention can create synaptic tags for the learning of working memories in sequential tasks.

Intelligence is our ability to learn appropriate responses to new stimuli and situations. Neurons in association cortex are thought to be essential for this ability. During learning these neurons become tuned to relevant features and start to represent them with persistent activity during memory delays. This learning process is not well understood. Here we develop a biologically plausible learning scheme that explains how trial-and-error learning induces neuronal selectivity and working memory representations for task-relevant information. We propose that the response selection stage sends attentional feedback signals to earlier processing levels, forming synaptic tags at those connections responsible for the stimulus-response mapping. Globally released neuromodulators then interact with tagged synapses to determine their plasticity. The resulting learning rule endows neural networks with the capacity to create new working memory representations of task relevant information as persistent activity. It is remarkably generic: it explains how association neurons learn to store task-relevant information for linear as well as non-linear stimulus-response mappings, how they become tuned to category boundaries or analog variables, depending on the task demands, and how they learn to integrate probabilistic evidence for perceptual decisions.

The deep latent space particle filter for real-time data assimilation with uncertainty quantification.

In data assimilation, observations are fused with simulations to obtain an accurate estimate of the state and parameters for a given physical system. Combining data with a model, however, while accurately estimating uncertainty, is computationally expensive and infeasible to run in real-time for complex systems. Here, we present a novel particle filter methodology, the Deep Latent Space Particle filter or D-LSPF, that uses neural network-based surrogate models to overcome this computational challenge. The D-LSPF enables filtering in the low-dimensional latent space obtained using Wasserstein AEs with modified vision transformer layers for dimensionality reduction and transformers for parameterized latent space time stepping. As we demonstrate on three test cases, including leak localization in multi-phase pipe flow and seabed identification for fully nonlinear water waves, the D-LSPF runs orders of magnitude faster than a high-fidelity particle filter and 3-5 times faster than alternative methods while being up to an order of magnitude more accurate. The D-LSPF thus enables real-time data assimilation with uncertainty quantification for the test cases demonstrated in this paper.

Predictive coding with spiking neurons and feedforward gist signaling.

Predictive coding (PC) is an influential theory in neuroscience, which suggests the existence of a cortical architecture that is constantly generating and updating predictive representations of sensory inputs. Owing to its hierarchical and generative nature, PC has inspired many computational models of perception in the literature. However, the biological plausibility of existing models has not been sufficiently explored due to their use of artificial neurons that approximate neural activity with firing rates in the continuous time domain and propagate signals synchronously. Therefore, we developed a spiking neural network for predictive coding (SNN-PC), in which neurons communicate using event-driven and asynchronous spikes. Adopting the hierarchical structure and Hebbian learning algorithms from previous PC neural network models, SNN-PC introduces two novel features: (1) a fast feedforward sweep from the input to higher areas, which generates a spatially reduced and abstract representation of input (i.e., a neural code for the gist of a scene) and provides a neurobiological alternative to an arbitrary choice of priors; and (2) a separation of positive and negative error-computing neurons, which counters the biological implausibility of a bi-directional error neuron with a very high baseline firing rate. After training with the MNIST handwritten digit dataset, SNN-PC developed hierarchical internal representations and was able to reconstruct samples it had not seen during training. SNN-PC suggests biologically plausible mechanisms by which the brain may perform perceptual inference and learning in an unsupervised manner. In addition, it may be used in neuromorphic applications that can utilize its energy-efficient, event-driven, local learning, and parallel information processing nature.

Local minimization of prediction errors drives learning of invariant object representations in a generative network model of visual perception.

The ventral visual processing hierarchy of the cortex needs to fulfill at least two key functions: perceived objects must be mapped to high-level representations invariantly of the precise viewing conditions, and a generative model must be learned that allows, for instance, to fill in occluded information guided by visual experience. Here, we show how a multilayered predictive coding network can learn to recognize objects from the bottom up and to generate specific representations via a top-down pathway through a single learning rule: the local minimization of prediction errors. Trained on sequences of continuously transformed objects, neurons in the highest network area become tuned to object identity invariant of precise position, comparable to inferotemporal neurons in macaques. Drawing on this, the dynamic properties of invariant object representations reproduce experimentally observed hierarchies of timescales from low to high levels of the ventral processing stream. The predicted faster decorrelation of error-neuron activity compared to representation neurons is of relevance for the experimental search for neural correlates of prediction errors. Lastly, the generative capacity of the network is confirmed by reconstructing specific object images, robust to partial occlusion of the inputs. By learning invariance from temporal continuity within a generative model, the approach generalizes the predictive coding framework to dynamic inputs in a more biologically plausible way than self-supervised networks with non-local error-backpropagation. This was achieved simply by shifting the training paradigm to dynamic inputs, with little change in architecture and learning rule from static input-reconstructing Hebbian predictive coding networks.

Streaming parallel GPU acceleration of large-scale filter-based spiking neural networks.

The arrival of graphics processing (GPU) cards suitable for massively parallel computing promises affordable large-scale neural network simulation previously only available at supercomputing facilities. While the raw numbers suggest that GPUs may outperform CPUs by at least an order of magnitude, the challenge is to develop fine-grained parallel algorithms to fully exploit the particulars of GPUs. Computation in a neural network is inherently parallel and thus a natural match for GPU architectures: given inputs, the internal state for each neuron can be updated in parallel. We show that for filter-based spiking neurons, like the Spike Response Model, the additive nature of membrane potential dynamics enables additional update parallelism. This also reduces the accumulation of numerical errors when using single precision computation, the native precision of GPUs. We further show that optimizing simulation algorithms and data structures to the GPU's architecture has a large pay-off: for example, matching iterative neural updating to the memory architecture of the GPU speeds up this simulation step by a factor of three to five. With such optimizations, we can simulate in better-than-realtime plausible spiking neural networks of up to 50 000 neurons, processing over 35 million spiking events per second.

Deep Gated Hebbian Predictive Coding Accounts for Emergence of Complex Neural Response Properties Along the Visual Cortical Hierarchy.

Predictive coding provides a computational paradigm for modeling perceptual processing as the construction of representations accounting for causes of sensory inputs. Here, we developed a scalable, deep network architecture for predictive coding that is trained using a gated Hebbian learning rule and mimics the feedforward and feedback connectivity of the cortex. After training on image datasets, the models formed latent representations in higher areas that allowed reconstruction of the original images. We analyzed low- and high-level properties such as orientation selectivity, object selectivity and sparseness of neuronal populations in the model. As reported experimentally, image selectivity increased systematically across ascending areas in the model hierarchy. Depending on the strength of regularization factors, sparseness also increased from lower to higher areas. The results suggest a rationale as to why experimental results on sparseness across the cortical hierarchy have been inconsistent. Finally, representations for different object classes became more distinguishable from lower to higher areas. Thus, deep neural networks trained using a gated Hebbian formulation of predictive coding can reproduce several properties associated with neuronal responses along the visual cortical hierarchy.

Multimodal Representation Learning for Place Recognition Using Deep Hebbian Predictive Coding.

Recognising familiar places is a competence required in many engineering applications that interact with the real world such as robot navigation. Combining information from different sensory sources promotes robustness and accuracy of place recognition. However, mismatch in data registration, dimensionality, and timing between modalities remain challenging problems in multisensory place recognition. Spurious data generated by sensor drop-out in multisensory environments is particularly problematic and often resolved through adhoc and brittle solutions. An effective approach to these problems is demonstrated by animals as they gracefully move through the world. Therefore, we take a neuro-ethological approach by adopting self-supervised representation learning based on a neuroscientific model of visual cortex known as predictive coding. We demonstrate how this parsimonious network algorithm which is trained using a local learning rule can be extended to combine visual and tactile sensory cues from a biomimetic robot as it naturally explores a visually aliased environment. The place recognition performance obtained using joint latent representations generated by the network is significantly better than contemporary representation learning techniques. Further, we see evidence of improved robustness at place recognition in face of unimodal sensor drop-out. The proposed multimodal deep predictive coding algorithm presented is also linearly extensible to accommodate more than two sensory modalities, thereby providing an intriguing example of the value of neuro-biologically plausible representation learning for multimodal navigation.

Visualizing a joint future of neuroscience and neuromorphic engineering.

Recent research resolves the challenging problem of building biophysically plausible spiking neural models that are also capable of complex information processing. This advance creates new opportunities in neuroscience and neuromorphic engineering, which we discussed at an online focus meeting.

Sparse Computation in Adaptive Spiking Neural Networks.

Artificial Neural Networks (ANNs) are bio-inspired models of neural computation that have proven highly effective. Still, ANNs lack a natural notion of time, and neural units in ANNs exchange analog values in a frame-based manner, a computationally and energetically inefficient form of communication. This contrasts sharply with biological neurons that communicate sparingly and efficiently using isomorphic binary spikes. While Spiking Neural Networks (SNNs) can be constructed by replacing the units of an ANN with spiking neurons (Cao et al., 2015; Diehl et al., 2015) to obtain reasonable performance, these SNNs use Poisson spiking mechanisms with exceedingly high firing rates compared to their biological counterparts. Here we show how spiking neurons that employ a form of neural coding can be used to construct SNNs that match high-performance ANNs and match or exceed state-of-the-art in SNNs on important benchmarks, while requiring firing rates compatible with biological findings. For this, we use spike-based coding based on the firing rate limiting adaptation phenomenon observed in biological spiking neurons. This phenomenon can be captured in fast adapting spiking neuron models, for which we derive the effective transfer function. Neural units in ANNs trained with this transfer function can be substituted directly with adaptive spiking neurons, and the resulting Adaptive SNNs (AdSNNs) can carry out competitive classification in deep neural networks without further modifications. Adaptive spike-based coding additionally allows for the dynamic control of neural coding precision: we show empirically how a simple model of arousal in AdSNNs further halves the average required firing rate and this notion naturally extends to other forms of attention as studied in neuroscience. AdSNNs thus hold promise as a novel and sparsely active model for neural computation that naturally fits to temporally continuous and asynchronous applications.

Visual pathways from the perspective of cost functions and multi-task deep neural networks.

Vision research has been shaped by the seminal insight that we can understand the higher-tier visual cortex from the perspective of multiple functional pathways with different goals. In this paper, we try to give a computational account of the functional organization of this system by reasoning from the perspective of multi-task deep neural networks. Machine learning has shown that tasks become easier to solve when they are decomposed into subtasks with their own cost function. We hypothesize that the visual system optimizes multiple cost functions of unrelated tasks and this causes the emergence of a ventral pathway dedicated to vision for perception, and a dorsal pathway dedicated to vision for action. To evaluate the functional organization in multi-task deep neural networks, we propose a method that measures the contribution of a unit towards each task, applying it to two networks that have been trained on either two related or two unrelated tasks, using an identical stimulus set. Results show that the network trained on the unrelated tasks shows a decreasing degree of feature representation sharing towards higher-tier layers while the network trained on related tasks uniformly shows high degree of sharing. We conjecture that the method we propose can be used to analyze the anatomical and functional organization of the visual system and beyond. We predict that the degree to which tasks are related is a good descriptor of the degree to which they share downstream cortical-units.

Adaptive resource allocation for efficient patient scheduling.

OBJECTIVE: Efficient scheduling of patient appointments on expensive resources is a complex and dynamic task. A resource is typically used by several patient groups. To service these groups, resource capacity is often allocated per group, explicitly or implicitly. Importantly, due to fluctuations in demand, for the most efficient use of resources this allocation must be flexible. METHODS: We present an adaptive approach to automatic optimization of resource calendars. In our approach, the allocation of capacity to different patient groups is flexible and adaptive to the current and expected future situation. We additionally present an approach to determine optimal resource openings hours on a larger time frame. Our model and its parameter values are based on extensive case analysis at the Academic Medical Hospital Amsterdam. RESULTS AND CONCLUSION: We have implemented a comprehensive computer simulation of the application case. Simulation experiments show that our approach of adaptive capacity allocation improves the performance of scheduling patients groups with different attributes and makes efficient use of resource capacity.