Graph neural networks (GNNs) have demonstrated strong performance on a wide variety of tasks due to their ability to model nonuniform structured data. Despite their promise, there exists little research exploring methods to make them more efficient at inference time. In this work, we explore the viability of training quantized GNNs, enabling the usage of low precision integer arithmetic during inference. We identify the sources of error that uniquely arise when attempting to quantize GNNs, and propose an architecturally-agnostic method, Degree-Quant, to improve performance over existing quantization-aware training baselines commonly used on other architectures, such as CNNs. We validate our method on six datasets and show, unlike previous attempts, that models generalize to unseen graphs. Models trained with Degree-Quant for INT8 quantization perform as well as FP32 models in most cases; for INT4 models, we obtain up to 26% gains over the baselines. Our work enables up to 4.7 speedups on CPU when using INT8 arithmetic. Graph neural networks (GNNs) have received substantial attention in recent years due to their ability to model irregularly structured data. As a result, they are extensively used for applications as diverse as molecular interactions (Duvenaud et al., 2015; Wu et al., 2017), social networks (Hamilton et al., 2017), recommendation systems (van den Berg et al., 2017) or program understanding (Allamanis et al., 2018). Recent advancements have centered around building more sophisticated models including new types of layers (Kipf & Welling, 2017; Velickovic et al., 2018; Xu et al., 2019) and better aggregation functions (Corso et al., 2020).

Despite the soaring use of convolutional neural networks (CNNs) in mobile applications, uniformly sustaining high-performance inference on mobile has been elusive due to the excessive computational demands of modern CNNs and the increasing diversity of deployed devices. A popular alternative comprises offloading CNN processing to powerful cloud-based servers. Nevertheless, by relying on the cloud to produce outputs, emerging mission-critical and high-mobility applications, such as drone obstacle avoidance or interactive applications, can suffer from the dynamic connectivity conditions and the uncertain availability of the cloud. In this paper, we propose SPINN, a distributed inference system that employs synergistic device-cloud computation together with a progressive inference method to deliver fast and robust CNN inference across diverse settings. The proposed system introduces a novel scheduler that co-optimises the early-exit policy and the CNN splitting at run time, in order to adapt to dynamic conditions and meet user-defined service-level requirements. Quantitative evaluation illustrates that SPINN outperforms its state-of-the-art collaborative inference counterparts by up to 2x in achieved throughput under varying network conditions, reduces the server cost by up to 6.8x and improves accuracy by 20.7% under latency constraints, while providing robust operation under uncertain connectivity conditions and significant energy savings compared to cloud-centric execution.

Increasing demand for on-device Automatic Speech Recognition (ASR) systems has resulted in renewed interests in developing automatic model compression techniques. Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7x, outperforming laborious manual rank-selection approaches. However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER). In this work, we propose an iterative AutoML-based LRF approach that achieves over 5x compression without degrading the WER, thereby advancing the state-of-the-art in ASR compression.

Federated Learning (FL) has emerged as a promising technique for edge devices to collaboratively learn a shared prediction model, while keeping their training data on the device, thereby decoupling the ability to do machine learning from the need to store the data in the cloud. However, FL is difficult to implement and deploy in practice, considering the heterogeneity in mobile devices, e.g., different programming languages, frameworks, and hardware accelerators. Although there are a few frameworks available to simulate FL algorithms (e.g., TensorFlow Federated), they do not support implementing FL workloads on mobile devices. Furthermore, these frameworks are designed to simulate FL in a server environment and hence do not allow experimentation in distributed mobile settings for a large number of clients. In this paper, we present Flower (https://flower.dev/), a FL framework which is both agnostic towards heterogeneous client environments and also scales to a large number of clients, including mobile and embedded devices. Flower's abstractions let developers port existing mobile workloads with little overhead, regardless of the programming language or ML framework used, while also allowing researchers flexibility to experiment with novel approaches to advance the state-of-the-art. We describe the design goals and implementation considerations of Flower and show our experiences in evaluating the performance of FL across clients with heterogeneous computational and communication capabilities.

Designing deep learning models for highly-constrained hardware would allow imbuing many edge devices with intelligence. Microcontrollers (MCUs) are an attractive platform for building smart devices due to their low cost, wide availability, and modest power usage. However, they lack the computational resources to run neural networks as straightforwardly as mobile or server platforms, which necessitates changes to the network architecture and the inference software. In this work, we discuss the deployment and memory concerns of neural networks on MCUs and present a way of saving memory by changing the execution order of the network's operators, which is orthogonal to other compression methods. We publish a tool for reordering operators of TensorFlow Lite models and demonstrate its utility by sufficiently reducing the memory footprint of a CNN to deploy it on an MCU with 512KB SRAM.

End-to-end automatic speech recognition (ASR) models are increasingly large and complex to achieve the best possible accuracy. In this paper, we build an AutoML system that uses reinforcement learning (RL) to optimize the per-layer compression ratios when applied to a state-of-the-art attention based end-to-end ASR model composed of several LSTM layers. We use singular value decomposition (SVD) low-rank matrix factorization as the compression method. For our RL-based AutoML system, we focus on practical considerations such as the choice of the reward/punishment functions, the formation of an effective search space, and the creation of a representative but small data set for quick evaluation between search steps. Finally, we present accuracy results on LibriSpeech of the model compressed by our AutoML system, and we compare it to manually-compressed models. Our results show that in the absence of retraining our RL-based search is an effective and practical method to compress a production-grade ASR system. When retraining is possible, we show that our AutoML system can select better highly-compressed seed models compared to manually hand-crafted rank selection, thus allowing for more compression than previously possible.

In recent years, advances in deep learning have resulted in unprecedented leaps in diverse tasks spanning from speech and object recognition to context awareness and health monitoring. As a result, an increasing number of AI-enabled applications are being developed targeting ubiquitous and mobile devices. While deep neural networks (DNNs) are getting bigger and more complex, they also impose a heavy computational and energy burden on the host devices, which has led to the integration of various specialized processors in commodity devices. Given the broad range of competing DNN architectures and the heterogeneity of the target hardware, there is an emerging need to understand the compatibility between DNN-platform pairs and the expected performance benefits on each platform. This work attempts to demystify this landscape by systematically evaluating a collection of state-of-the-art DNNs on a wide variety of commodity devices. In this respect, we identify potential bottlenecks in each architecture and provide important guidelines that can assist the community in the co-design of more efficient DNNs and accelerators.

For people with chronic pain (CP), the assessment of protective behavior during physical functioning is essential to understand their subjective pain-related experiences (e.g., fear and anxiety toward pain and injury) and how they deal with such experiences (avoidance or reliance on specific body joints), with the ultimate goal of guiding intervention. Advances in deep learning (DL) can enable the development of such intervention. Using the EmoPain MoCap dataset, we investigate how attention-based DL architectures can be used to improve the detection of protective behavior by capturing the most informative biomechanical cues characterizing specific movements and the strategies used to execute them to cope with pain-related experience. We propose an end-to-end neural network architecture based on attention mechanism, named BodyAttentionNet (BANet). BANet is designed to learn temporal and body-joint regions that are informative to the detection of protective behavior. The approach can consider the variety of ways people execute one movement (including healthy people) and it is independent of the type of movement analyzed. We also explore variants of this architecture to understand the contribution of both temporal and bodily attention mechanisms. Through extensive experiments with other state-of-the-art machine learning techniques used with motion capture data, we show a statistically significant improvement achieved by combining the two attention mechanisms. In addition, the BANet architecture requires a much lower number of parameters than the state-of-the-art ones for comparable if not higher performances.

In chronic pain physical rehabilitation, physiotherapists adapt movement to current performance of patients especially based on the expression of protective behavior, gradually exposing them to feared but harmless and essential everyday movements. As physical rehabilitation moves outside the clinic, physical rehabilitation technology needs to automatically detect such behaviors so as to provide similar personalized support. In this paper, we investigate the use of a Long Short-Term Memory (LSTM) network, which we call Protect-LSTM, to detect events of protective behavior, based on motion capture and electromyography data of healthy people and people with chronic low back pain engaged in five everyday movements. Differently from previous work on the same dataset, we aim to continuously detect protective behavior within a movement rather than overall estimate the presence of such behavior. The Protect-LSTM reaches best average F1 score of 0.815 with leave-one-subject-out (LOSO) validation, using low level features, better than other algorithms. Performances increase for some movements when modelled separately (mean F1 scores: bending=0.77, standing on one leg=0.81, sit-to-stand=0.72, stand-to-sit=0.83, reaching forward=0.67). These results reach excellent level of agreement with the average ratings of physiotherapists. As such, the results show clear potential for in-home technology supported affect-based personalized physical rehabilitation.

We analyse multimodal time-series data corresponding to weight, sleep and steps measurements. We focus on predicting whether a user will successfully achieve his/her weight objective. For this, we design several deep long short-term memory (LSTM) architectures, including a novel cross-modal LSTM (X-LSTM), and demonstrate their superiority over baseline approaches. The X-LSTM improves parameter efficiency by processing each modality separately and allowing for information flow between them by way of recurrent cross-connections. We present a general hyperparameter optimisation technique for X-LSTMs, which allows us to significantly improve on the LSTM and a prior state-of-the-art cross-modal approach, using a comparable number of parameters. Finally, we visualise the model's predictions, revealing implications about latent variables in this task.