Design Space Exploration of Dense and Sparse Mapping Schemes for RRAM Architectures

The impact of device and circuit-level effects in mixed-signal Resistive Random Access Memory (RRAM) accelerators typically manifest as performance degradation of Deep Learning (DL) algorithms, but the degree of impact varies based on algorithmic features. These include network architecture, capacity, weight distribution, and the type of inter-layer connections. Techniques are continuously emerging to efficiently train sparse neural networks, which may have activation sparsity, quantization, and memristive noise. In this paper, we present an extended Design Space Exploration (DSE) methodology to quantify the benefits and limitations of dense and sparse mapping schemes for a variety of network architectures. While sparsity of connectivity promotes less power consumption and is often optimized for extracting localized features, its performance on tiled RRAM arrays may be more susceptible to noise due to under-parameterization, when compared to dense mapping schemes. Moreover, we present a case study quantifying and formalizing the trade-offs of typical non-idealities introduced into 1-Transistor-1-Resistor (1T1R) tiled memristive architectures and the size of modular crossbar tiles using the CIFAR-10 dataset.

A Deep Learning Localization Method for Measuring Abdominal Muscle Dimensions in Ultrasound Images

Health professionals extensively use Two- Dimensional (2D) Ultrasound (US) videos and images to visualize and measure internal organs for various purposes including evaluation of muscle architectural changes. US images can be used to measure abdominal muscles dimensions for the diagnosis and creation of customized treatment plans for patients with Low Back Pain (LBP), however, they are difficult to interpret. Due to high variability, skilled professionals with specialized training are required to take measurements to avoid low intra-observer reliability. This variability stems from the challenging nature of accurately finding the correct spatial location of measurement endpoints in abdominal US images. In this paper, we use a Deep Learning (DL) approach to automate the measurement of the abdominal muscle thickness in 2D US images. By treating the problem as a localization task, we develop a modified Fully Convolutional Network (FCN) architecture to generate blobs of coordinate locations of measurement endpoints, similar to what a human operator does. We demonstrate that using the TrA400 US image dataset, our network achieves a Mean Absolute Error (MAE) of 0.3125 on the test set, which almost matches the performance of skilled ultrasound technicians. Our approach can facilitate next steps for automating the process of measurements in 2D US images, while reducing inter-observer as well as intra-observer variability for more effective clinical outcomes.

Memristive Stochastic Computing for Deep Learning Parameter Optimization

Stochastic Computing (SC) is a computing paradigm that allows for the low-cost and low-power computation of various arithmetic operations using stochastic bit streams and digital logic. In contrast to conventional representation schemes used within the binary domain, the sequence of bit streams in the stochastic domain is inconsequential, and computation is usually non-deterministic. In this brief, we exploit the stochasticity during switching of probabilistic Conductive Bridging RAM (CBRAM) devices to efficiently generate stochastic bit streams in order to perform Deep Learning (DL) parameter optimization, reducing the size of Multiply and Accumulate (MAC) units by 5 orders of magnitude. We demonstrate that in using a 40-nm Complementary Metal Oxide Semiconductor (CMOS) process our scalable architecture occupies 1.55mm$^2$ and consumes approximately 167$\mu$W when optimizing parameters of a Convolutional Neural Network (CNN) while it is being trained for a character recognition task, observing no notable reduction in accuracy post-training.

Affinity LCFCN: Learning to Segment Fish with Weak Supervision

Aquaculture industries rely on the availability of accurate fish body measurements, e.g., length, width and mass. Manual methods that rely on physical tools like rulers are time and labour intensive. Leading automatic approaches rely on fully-supervised segmentation models to acquire these measurements but these require collecting per-pixel labels -- also time consuming and laborious: i.e., it can take up to two minutes per fish to generate accurate segmentation labels, almost always requiring at least some manual intervention. We propose an automatic segmentation model efficiently trained on images labeled with only point-level supervision, where each fish is annotated with a single click. This labeling process requires significantly less manual intervention, averaging roughly one second per fish. Our approach uses a fully convolutional neural network with one branch that outputs per-pixel scores and another that outputs an affinity matrix. We aggregate these two outputs using a random walk to obtain the final, refined per-pixel segmentation output. We train the entire model end-to-end with an LCFCN loss, resulting in our A-LCFCN method. We validate our model on the DeepFish dataset, which contains many fish habitats from the north-eastern Australian region. Our experimental results confirm that A-LCFCN outperforms a fully-supervised segmentation model at fixed annotation budget. Moreover, we show that A-LCFCN achieves better segmentation results than LCFCN and a standard baseline. We have released the code at \url{https://github.com/IssamLaradji/affinity_lcfcn}.

Hardware Implementation of Deep Network Accelerators Towards Healthcare and Biomedical Applications

With the advent of dedicated Deep Learning (DL) accelerators and neuromorphic processors, new opportunities are emerging for applying deep and Spiking Neural Network (SNN) algorithms to healthcare and biomedical applications at the edge. This can facilitate the advancement of the medical Internet of Things (IoT) systems and Point of Care (PoC) devices. In this paper, we provide a tutorial describing how various technologies ranging from emerging memristive devices, to established Field Programmable Gate Arrays (FPGAs), and mature Complementary Metal Oxide Semiconductor (CMOS) technology can be used to develop efficient DL accelerators to solve a wide variety of diagnostic, pattern recognition, and signal processing problems in healthcare. Furthermore, we explore how spiking neuromorphic processors can complement their DL counterparts for processing biomedical signals. After providing the required background, we unify the sparsely distributed research on neural network and neuromorphic hardware implementations as applied to the healthcare domain. In addition, we benchmark various hardware platforms by performing a biomedical electromyography (EMG) signal processing task and drawing comparisons among them in terms of inference delay and energy. Finally, we provide our analysis of the field and share a perspective on the advantages, disadvantages, challenges, and opportunities that different accelerators and neuromorphic processors introduce to healthcare and biomedical domains. This paper can serve a large audience, ranging from nanoelectronics researchers, to biomedical and healthcare practitioners in grasping the fundamental interplay between hardware, algorithms, and clinical adoption of these tools, as we shed light on the future of deep networks and spiking neuromorphic processing systems as proponents for driving biomedical circuits and systems forward.

Training Progressively Binarizing Deep Networks Using FPGAs

While hardware implementations of inference routines for Binarized Neural Networks (BNNs) are plentiful, current realizations of efficient BNN hardware training accelerators, suitable for Internet of Things (IoT) edge devices, leave much to be desired. Conventional BNN hardware training accelerators perform forward and backward propagations with parameters adopting binary representations, and optimization using parameters adopting floating or fixed-point real-valued representations--requiring two distinct sets of network parameters. In this paper, we propose a hardware-friendly training method that, contrary to conventional methods, progressively binarizes a singular set of fixed-point network parameters, yielding notable reductions in power and resource utilizations. We use the Intel FPGA SDK for OpenCL development environment to train our progressively binarizing DNNs on an OpenVINO FPGA. We benchmark our training approach on both GPUs and FPGAs using CIFAR-10 and compare it to conventional BNNs.

Variation-aware Binarized Memristive Networks

The quantization of weights to binary states in Deep Neural Networks (DNNs) can replace resource-hungry multiply accumulate operations with simple accumulations. Such Binarized Neural Networks (BNNs) exhibit greatly reduced resource and power requirements. In addition, memristors have been shown as promising synaptic weight elements in DNNs. In this paper, we propose and simulate novel Binarized Memristive Convolutional Neural Network (BMCNN) architectures employing hybrid weight and parameter representations. We train the proposed architectures offline and then map the trained parameters to our binarized memristive devices for inference. To take into account the variations in memristive devices, and to study their effect on the performance, we introduce variations in $R_{ON}$ and $R_{OFF}$. Moreover, we introduce means to mitigate the adverse effect of memristive variations in our proposed networks. Finally, we benchmark our BMCNNs and variation-aware BMCNNs using the MNIST dataset.

Accelerating Deterministic and Stochastic Binarized Neural Networks on FPGAs Using OpenCL

Recent technological advances have proliferated the available computing power, memory, and speed of modern Central Processing Units (CPUs), Graphics Processing Units (GPUs), and Field Programmable Gate Arrays (FPGAs). Consequently, the performance and complexity of Artificial Neural Networks (ANNs) is burgeoning. While GPU accelerated Deep Neural Networks (DNNs) currently offer state-of-the-art performance, they consume large amounts of power. Training such networks on CPUs is inefficient, as data throughput and parallel computation is limited. FPGAs are considered a suitable candidate for performance critical, low power systems, e.g. the Internet of Things (IOT) edge devices. Using the Xilinx SDAccel or Intel FPGA SDK for OpenCL development environment, networks described using the high-level OpenCL framework can be accelerated on heterogeneous platforms. Moreover, the resource utilization and power consumption of DNNs can be further enhanced by utilizing regularization techniques that binarize network weights. In this paper, we introduce, to the best of our knowledge, the first FPGA-accelerated stochastically binarized DNN implementations, and compare them to implementations accelerated using both GPUs and FPGAs. Our developed networks are trained and benchmarked using the popular MNIST and CIFAR-10 datasets, and achieve near state-of-the-art performance, while offering a >16-fold improvement in power consumption, compared to conventional GPU-accelerated networks. Both our FPGA-accelerated determinsitic and stochastic BNNs reduce inference times on MNIST and CIFAR-10 by >9.89x and >9.91x, respectively.

DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning

Robotic weed control has seen increased research in the past decade with its potential for boosting productivity in agriculture. Majority of works focus on developing robotics for arable croplands, ignoring the significant weed management problems facing rangeland stock farmers. Perhaps the greatest obstacle to widespread uptake of robotic weed control is the robust detection of weed species in their natural environment. The unparalleled successes of deep learning make it an ideal candidate for recognising various weed species in the highly complex Australian rangeland environment. This work contributes the first large, public, multiclass image dataset of weed species from the Australian rangelands; allowing for the development of robust detection methods to make robotic weed control viable. The DeepWeeds dataset consists of 17,509 labelled images of eight nationally significant weed species native to eight locations across northern Australia. This paper also presents a baseline for classification performance on the dataset using the benchmark deep learning models, Inception-v3 and ResNet-50. These models achieved an average classification performance of 87.9% and 90.5%, respectively. This strong result bodes well for future field implementation of robotic weed control methods in the Australian rangelands.

A Neuromorphic VLSI Design for Spike Timing and Rate Based Synaptic Plasticity

Triplet-based Spike Timing Dependent Plasticity (TSTDP) is a powerful synaptic plasticity rule that acts beyond conventional pair-based STDP (PSTDP). Here, the TSTDP is capable of reproducing the outcomes from a variety of biological experiments, while the PSTDP rule fails to reproduce them. Additionally, it has been shown that the behaviour inherent to the spike rate-based Bienenstock-Cooper-Munro (BCM) synaptic plasticity rule can also emerge from the TSTDP rule. This paper proposes an analog implementation of the TSTDP rule. The proposed VLSI circuit has been designed using the AMS 0.35 um CMOS process and has been simulated using design kits for Synopsys and Cadence tools. Simulation results demonstrate how well the proposed circuit can alter synaptic weights according to the timing difference amongst a set of different patterns of spikes. Furthermore, the circuit is shown to give rise to a BCM-like learning rule, which is a rate-based rule. To mimic implementation environment, a 1000 run Monte Carlo (MC) analysis was conducted on the proposed circuit. The presented MC simulation analysis and the simulation result from fine-tuned circuits show that, it is possible to mitigate the effect of process variations in the proof of concept circuit, however, a practical variation aware design technique is required to promise a high circuit performance in a large scale neural network. We believe that the proposed design can play a significant role in future VLSI implementations of both spike timing and rate based neuromorphic learning systems.