In visual scene understanding tasks, it is essential to capture both invariant and equivariant structure. While neural networks are frequently trained to achieve invariance to transformations such as translation, this often comes at the cost of losing access to equivariant information - e.g., the precise location of an object. Moreover, invariance is not naturally guaranteed through supervised learning alone, and many architectures generalize poorly to input transformations not encountered during training. Here, we take an approach based on analysis-by-synthesis and factoring using resonator networks. A generative model describes the construction of simple scenes containing MNIST digits and their transformations, like color and position. The resonator network inverts the generative model, and provides both invariant and equivariant information about particular objects. Sparse features learned from training data act as a basis set to provide flexibility in representing variable shapes of objects, allowing the resonator network to handle previously unseen digit shapes from the test set. The modular structure provides a shape module which contains information about the object shape with translation factored out, allowing a simple classifier to operate on centered digits. The classification layer is trained solely on centered data, requiring much less training data, and the network as a whole can identify objects with arbitrary translations without data augmentation. The natural attention-like mechanism of the resonator network also allows for analysis of scenes with multiple objects, where the network dynamics selects and centers only one object at a time. Further, the specific position information of a particular object can be extracted from the translation module, and we show that the resonator can be designed to track multiple moving objects with precision of a few pixels.

Analysing a visual scene by inferring the configuration of a generative model is widely considered the most flexible and generalizable approach to scene understanding. Yet, one major problem is the computational challenge of the inference procedure, involving a combinatorial search across object identities and poses. Here we propose a neuromorphic solution exploiting three key concepts: (1) a computational framework based on vector symbolic architectures (VSAs) with complex-valued vectors, (2) the design of hierarchical resonator networks to factorize the non-commutative transforms translation and rotation in visual scenes and (3) the design of a multi-compartment spiking phasor neuron model for implementing complex-valued resonator networks on neuromorphic hardware. The VSA framework uses vector binding operations to form a generative image model in which binding acts as the equivariant operation for geometric transformations. A scene can therefore be described as a sum of vector products, which can then be efficiently factorized by a resonator network to infer objects and their poses. The hierarchical resonator network features a partitioned architecture in which vector binding is equivariant for horizontal and vertical translation within one partition and for rotation and scaling within the other partition. The spiking neuron model allows mapping the resonator network onto efficient and low-power neuromorphic hardware. Our approach is demonstrated on synthetic scenes composed of simple two-dimensional shapes undergoing rigid geometric transformations and colour changes. A companion paper demonstrates the same approach in real-world application scenarios for machine vision and robotics. The inference procedure for analysing a visual scene presents a computational challenge. Renner, Supic and colleagues develop a neural network model, the hierarchical resonator, to determine the generative factors of variation of objects in simple scenes. The resonator was implemented on neuromorphic hardware, using a spike-timing code for complex numbers.

Visual odometry (VO) is a method used to estimate self-motion of a mobile robot using visual sensors. Unlike odometry based on integrating differential measurements that can accumulate errors, such as inertial sensors or wheel encoders, VO is not compromised by drift. However, image-based VO is computationally demanding, limiting its application in use cases with low-latency, low-memory and low-energy requirements. Neuromorphic hardware offers low-power solutions to many vision and artificial intelligence problems, but designing such solutions is complicated and often has to be assembled from scratch. Here we propose the use of vector symbolic architecture (VSA) as an abstraction layer to design algorithms compatible with neuromorphic hardware. Building from a VSA model for scene analysis, described in our companion paper, we present a modular neuromorphic algorithm that achieves state-of-the-art performance on two-dimensional VO tasks. Specifically, the proposed algorithm stores and updates a working memory of the presented visual environment. Based on this working memory, a resonator network estimates the changing location and orientation of the camera. We experimentally validate the neuromorphic VSA-based approach to VO with two benchmarks: one based on an event-camera dataset and the other in a dynamic scene with a robotic task. Visual odometry, or self-motion estimation, is a fundamental task in robotics. Renner, Supic and colleagues introduce a neuromorphic algorithm for visual odometry that leverages hyperdimensional computing and hierarchical resonators. The approach estimates a robot's motion from event-based vision, a step towards low-power machine vision for robotics.

Beyond providing accurate movements, achieving smooth motion trajectories is a long-standing goal of robotics control theory for arms aiming to replicate natural human movements. Drawing inspiration from biological agents, whose reaching control networks effortlessly give rise to smooth and precise movements, can simplify these control objectives for robot arms. Neuromorphic processors, which mimic the brain’s computational principles, are an ideal platform to approximate the accuracy and smoothness of biological controllers while maximizing their energy efficiency and robustness. However, the incompatibility of conventional control methods with neuromorphic hardware limits the computational efficiency and explainability of their existing adaptations. In contrast, the neuronal subnetworks underlying smooth and accurate reaching movements are effective, minimal, and inherently compatible with neuromorphic hardware. In this work, we emulate these networks with a biologically realistic spiking neural network for motor control on neuromorphic hardware. The proposed controller incorporates experimentally-identified short-term synaptic plasticity and specialized neurons that regulate sensory feedback gain to provide smooth and accurate joint control across a wide motion range. Concurrently, it preserves the minimal complexity of its biological counterpart and is directly deployable on Intel’s neuromorphic processor. Using the joint controller as a building block and inspired by joint coordination in human arms, we scaled up this approach to control real-world robot arms. The trajectories and smooth, bell-shaped velocity profiles of the resulting motions resembled those of humans, verifying the biological relevance of the controller. Notably, the method achieved state-of-the-art control performance while decreasing the motion jerk by 19% to improve motion smoothness. Overall, this work suggests that control solutions inspired by experimentally identified neuronal architectures can provide effective, neuromorphic-controlled robots.

Elishai Ezra Tsur, a multidisciplinary researcher, talks about the challenges that conventional academic mindset brought to his professional life. He, DeWolf, and Supic introduce us with their viewpoint about "data science" and its role in their research. In their recent work published in this issue of Patterns, they tackle the inverse kinematics problem using brain-inspired neuronal architectures.

Inverse kinematics is fundamental for computational motion planning. It is used to derive an appropriate state in a robot's configuration space, given a target position in task space. In this work, we investigate the performance of fully connected and residual artificial neural networks as well as recurrent, learning-based, and deep spiking neural networks for conventional and geometrically constrained inverse kinematics. We show that while highly parameterized data-driven neural networks with tens to hundreds of thousands of parameters exhibit sub-ms inference time and sub-mm accuracy, learning-based spiking architectures can provide reasonably good results with merely a few thousand neurons. Moreover, we show that spiking neural networks can perform well in geometrically constrained task space, even when configured to an energy-conserved spiking rate, demonstrating their robustness. Neural networks were evaluated on NVIDIA's Xavier and Intel's neuromorphic Loihi chip.

In this paper, we explore the ability of a robot arm to learn the underlying operation space defined by the positions (x, y, z) that the arm's end-effector can reach, including disturbances, by deploying and thoroughly evaluating a Spiking Neural Network SNN-based adaptive control algorithm. While traditional control algorithms for robotics have limitations in both adapting to new and dynamic environments, we show that the robot arm can learn the operational space and complete tasks faster over time. We also demonstrate that the adaptive robot control algorithm based on SNNs enables a fast response while maintaining energy efficiency. We obtained these results by performing an extensive search of the adaptive algorithm parameter space, and evaluating algorithm performance for different SNN network sizes, learning rates, dynamic robot arm trajectories, and response times. We show that the robot arm learns to complete tasks 15% faster in specific experiment scenarios such as scenarios with six or nine random target points.

Wheelchair-mounted robotic arms support people with upper extremity disabilities with various activities of daily living (ADL). However, the associated cost and the power consumption of responsive and adaptive assistive robotic arms contribute to the fact that such systems are in limited use. Neuromorphic spiking neural networks can be used for a real-time machine learning-driven control of robots, providing an energy efficient framework for adaptive control. In this work, we demonstrate a neuromorphic adaptive control of a wheelchair-mounted robotic arm deployed on Intel’s Loihi chip. Our algorithm design uses neuromorphically represented and integrated velocity readings to derive the arm’s current state. The proposed controller provides the robotic arm with adaptive signals, guiding its motion while accounting for kinematic changes in real-time. We pilot-tested the device with an able-bodied participant to evaluate its accuracy while performing ADL-related trajectories. We further demonstrated the capacity of the controller to compensate for unexpected inertia-generating payloads using online learning. Videotaped recordings of ADL tasks performed by the robot were viewed by caregivers; data summarizing their feedback on the user experience and the potential benefit of the system is reported.

Neuromorphic implementation of robotic control has been shown to outperform conventional control paradigms in terms of robustness to perturbations and adaptation to varying conditions. Two main ingredients of robotics are inverse kinematic and Proportional–Integral–Derivative (PID) control. Inverse kinematics is used to compute an appropriate state in a robot's configuration space, given a target position in task space. PID control applies responsive correction signals to a robot's actuators, allowing it to reach its target accurately. The Neural Engineering Framework (NEF) offers a theoretical framework for a neuromorphic encoding of mathematical constructs with spiking neurons for the implementation of functional large-scale neural networks. In this work, we developed NEF-based neuromorphic algorithms for inverse kinematics and PID control, which we used to manipulate 6 degrees of freedom robotic arm. We used online learning for inverse kinematics and signal integration and differentiation for PID, offering high performing and energy-efficient neuromorphic control. Algorithms were evaluated in simulation as well as on Intel's Loihi neuromorphic hardware.

While commercial mid-air gesture recognition systems have existed for at least a decade, they have not become a widespread method of interacting with machines. This is primarily due to the fact that these systems require rigid, dramatic gestures to be performed for accurate recognition that can be fatiguing and unnatural. The global pandemic has seen a resurgence of interest in touchless interfaces, so new methods that allow for natural mid-air gestural inter-actions are even more important. To address the limitations of recognition systems, we propose a neuromorphic gesture analysis system which naturally declutters the background and analyzes gestures at high temporal resolution. Our novel model consists of an event-based guided Variational Autoencoder (VAE) which encodes event-based data sensed by a Dynamic Vision Sensor (DVS) into a latent space representation suitable to analyze and compute the similarity of mid-air gesture data. Our results show that the features learned by the VAE provides a similarity measure capable of clustering and pseudo labeling of new gestures. Furthermore, we argue that the resulting event-based encoder and pseudo-labeling system are suitable for implementation in neuromorphic hardware for online adaptation and learning of natural mid-air gestures.

A surge in artificial intelligence and autonomous technologies have increased the demand toward enhanced edge-processing capabilities. Computational complexity and size of state-of-the-art Deep Neural Networks (DNNs) are rising exponentially with diverse network models and larger datasets. This growth limits the performance scaling and energy-efficiency of both distributed and embedded inference platforms. Embedded designs at the edge are constrained by energy and speed limitations of available processor substrates and processor to memory communication required to fetch the model coefficients. While many hardware accelerator and network deployment frameworks have been in development, a framework is needed to allow the variety of existing architectures, and those in development, to be expressed in critical parts of the flow that perform various optimization steps. Moreover, premature architecture-blind network selection and optimization diminish the effectiveness of schedule optimizations and hardware-specific mappings. In this paper, we address these issues by creating a cross-layer software-hardware design framework that encompasses network training and model compression that is aware of and tuned to the underlying hardware architecture. This approach leverages the available degrees of DNN structure and sparsity to create a converged network that can be partitioned and efficiently scheduled on the target hardware platform, minimizing data movement, and improving the overall throughput and energy. To further streamline the design, we leverage the high-level, flexible SoC generator platform based on RISC-V ROCC framework. This integration allows seamless extensions of the RISC-V instruction set and Chisel-based rapid generator design. Utilizing this approach, we implemented a silicon prototype in a 16 nm TSMC process node achieving record processing efficiency of up to 18 TOPS/W.

Deep neural networks (DNNs) have become the state-of-the-art technique for machine learning tasks in various applications. However, due to their size and the computational complexity, large DNNs are not readily deployable on edge devices in real-time. To manage complexity and accelerate computation, network compression techniques based on pruning and quantization have been proposed and shown to be effective in reducing network size. However, such network compression can result in irregular matrix structures that are mismatched with modern hardware-accelerated platforms, such as graphics processing units (GPUs) designed to perform the DNN matrix multiplications in a structured (block-based) way. We propose MPDCompress, a DNN compression algorithm based on matrix permutation decomposition via random mask generation. In-training application of the masks molds the synaptic weight connection matrix to a sub-graph separation format. Aided by the random permutations, a hardware-desirable block matrix is generated, allowing for a more efficient implementation and compression of the network. To show versatility, we empirically verify MPDCompress on several network models, compression rates, and image datasets. On the LeNet 300-100 model (MNIST dataset), Deep MNIST, and CIFAR10, we achieve 10 X network compression with less than 1% accuracy loss compared to non-compressed accuracy performance. On AlexNet for the full ImageNet ILSVRC-2012 dataset, we achieve 8 X network compression with less than 1% accuracy loss, with top-5 and top-1 accuracies of 79.6% and 56.4%, respectively. Finally, we observe that the algorithm can offer inference speedups across various hardware platforms, with 4 X faster operation achieved on several mobile GPUs.

Deep Neural Networks (DNNs) are the key to the state-of-the-art machine vision, sensor fusion and audio/video signal processing. Unfortunately, their computation complexity and tight resource constraints on the Edge make them hard to leverage on mobile, embedded and IoT devices. Due to great diversity of Edge devices, DNN designers have to take into account the hardware platform and application requirements during network training. In this work we introduce pruning via matrix pivoting as a way to improve network pruning by compromising between the design flexibility of architecture-oblivious and performance efficiency of architecture-aware pruning, the two dominant techniques for obtaining resource-efficient DNNs. We also describe local and global network optimization techniques for efficient implementation of the resulting pruned networks. In combination, the proposed pruning and implementation result in close to linear speed up with the reduction of network coefficients during pruning.

This case study presents realization of one small grid tied self-consumption PV system within commercial building that uses AC coupled inverter and battery storage. Controlled production of energy makes MPPT controller, controlled by hybrid inverter through VE. Can communication network. The paper presents advantages of self-consumption battery storage systems, summarizes results on performance and lessons learned from realized self-consumption system of power 2,5kW.

The massive earthquake and tsunami off the coast of Japan on March 11, 2011 caused extensive damage at the Fukushima Daiichi nuclear power plant. During subsequent venting and explosions at the reactor site, there were releases of fission products such as 131I, 134Cs, 136Cs, 137Cs, and 132Te. Trace amounts of these isotopes were detectable in California around March 17. In the days after the disaster, the Berkeley Radiological Air and Water Monitoring (BRAWM) Project was started to measure the amounts of radioisotopes in the local environment around Berkeley. BRAWM has detected radioactive isotopes from Fukushima Daiichi in the air, rainwater, creek runoff, milk, soil, berries, and leafy vegetables. The team continues to monitor fallout levels in order to understand the nature of the radioactive releases from Fukushima as well as quantify the dilution or accumulation of the radioisotopes as they make their way through the environment and food chain.