FedZKT enables resource-constrained devices to participate in federated learning with heterogeneous models.
problem Inequality in resource allocation hinders participation from resource-constrained devices in federated learning.
method Zero-shot knowledge transfer through a server-assigned distillation process.
result FedZKT effectively transfers knowledge across heterogeneous on-device models without requiring comparable local training efforts.
This work reduces computation cost for on-device CNN training.
problem High computation cost during on-device CNN training.
method Self-supervised instance filtering and error map pruning.
result Substantial computation saving without significant accuracy loss.
This paper optimizes AI inference on edge devices with reduced communication and computation costs.
problem Efficiently performing AI inference on resource-constrained edge devices with reduced communication and computation costs.
method A three-step framework for effective inference: model split point selection, communication-aware model compression, and task-oriented encoding of intermediate features.
result Our proposed framework achieves a better trade-off and significantly reduces inference latency compared to baseline methods.
The paper optimizes neural network inference on mobile GPUs.
problem Limited computing power and thermal constraints on mobile CPUs.
method Leverage mobile GPUs for neural network inference.
result Real-time inference of deep neural networks on Android and iOS devices.
Federated learning framework extended for personalizing global models.
problem Evaluate personalization strategies for global models on-device.
method Extend federated learning framework, develop tools for analysis and evaluation.
result Personalization yields significant benefits for a large user population.
Survey on-device ML challenges and future directions.
problem Training machine learning models on-device with limited resources.
method Reformulated as resource constrained learning, comparing techniques from various AI areas.
result Identification of open challenges and future research directions.
On-device federated learning updates edge models by exchanging trained results.
problem Limited training data at edge devices due to model drift.
method OS-ELM for sequential training and autoencoder for anomaly detection, combined with federated learning.
result The proposed approach produces a merged model as accurately as traditional methods with lower costs.
Mobile training improves speech recognition for users with unique speech characteristics.
problem Limited generalization of speaker-independent speech recognition models for users with very different speech characteristics.
method Securely training personalized end-to-end speech recognition models on mobile devices, splitting gradient computation to reduce memory usage.
result On-device personalization achieved 58.1% relative word error rate reduction compared to 63.7% in a server environment, with 18.7% performance degradation.
This paper tackles energy-efficient machine learning on low-power devices.
problem Energy consumption in machine learning due to data communication.
method Dynamic averaging for integer exponential families on low-power processors.
result Achieves comparable model quality with significantly less communication and energy.
Much of the focus in the design of deep neural networks has been on improving accuracy, leading to more powerful yet highly complex network architectures that are difficult to deploy in practical scenarios, particularly on edge devices such as mobile and other consumer devices given their high computational and memory …
The rising popularity of intelligent mobile devices and the daunting computational cost of deep learning-based models call for efficient and accurate on-device inference schemes. We propose a quantization scheme that allows inference to be carried out using integer-only arithmetic, which can be implemented more efficie…
Personalized stress model using transfer learning from 20 participants.
problem Limited generalizability of machine learning models due to individual physiological differences.
method Transfer learning from a base model trained on 20 participants' physiological data collected in real-time.
result Improved model personalization and cross-domain performance.
Various health-care applications such as assisted living, fall detection etc., require modeling of user behavior through Human Activity Recognition (HAR). HAR using mobile- and wearable-based deep learning algorithms have been on the rise owing to the advancements in pervasive computing. However, there are two other ch…
We consider distributed on-device learning with limited communication and security requirements. We propose a new robust distributed optimization algorithm with efficient communication and attack tolerance. The proposed algorithm has provable convergence and robustness under non-IID settings. Empirical results show tha…
User-specific KWS system learns new keywords on-device.
problem Out-of-vocabulary problem in traditional KWS systems.
method Query-by-example enrollment and testing, using phonetic posteriors and FST.
result Promising performance on two keywords, preserving simplicity.
A new quantization strategy reduces Transformer model size and inference time.
problem Heavy computation load and memory overhead in Transformer models for mobile devices.
method Mixed precision quantization with varying bits per word in embedding blocks.
result 11.8x smaller model size and 3.5x speed up for on-device NMT.
On-device CNN inference for real-time computer vision applications can result in computational demands that far exceed the energy budgets of mobile devices. This paper proposes FixyNN, a co-designed hardware accelerator platform which splits a CNN model into two parts: a set of layers that are fixed in the hardware pla…
Paper proposes an AutoML framework for efficient device-edge co-inference.
problem Finding optimal hyper-parameters for model sparsity and feature compression.
method Sequential decision problem solved using deep reinforcement learning (DRL).
result Achieves better communication-computation trade-off and significant speedup.
A lightweight FPGA-based reinforcement learning approach for edge devices.
problem Resource constraints and inefficiency of DQN on edge devices.
method OS-ELM based training algorithm and L2 regularization for stability.
result 29.77x and 89.40x faster than conventional DQN-based approach for CartPole-v0 task.
A hybrid framework reduces ML complexity on edge devices.
problem Limited memory and energy on edge devices.
method Compressed data collection and tailored deep learning network.
result Significant reduction in computational complexity and memory.
Paper proposes federated learning for SNNs to enable low-power, online training.
problem Limited data at each device for on-device SNN training.
method Federated Learning (FL) for cooperative SNN training, leveraging local and global feedback.
result FL-SNN achieves significant advantages over separate training and offers a flexible trade-off between accuracy and communication load.
VoiceFilter-Lite separates speech from background in real-time for on-device speech recognition.
problem Separate speech from background in real-time for on-device speech recognition.
method Asymmetric loss, adaptive runtime suppression, quantization to 8-bit.
result VoiceFilter-Lite achieves real-time speech separation and maintains speech recognition performance.
Paper proposes low-rank gradient approximation to save memory for deep neural network training.
problem Memory limitation on mobile devices for deep neural network training.
method Approximating gradient matrices using low-rank parameterization.
result Reduces training memory by about 33.0% for Adam optimization and 4.5% relative lower word error rate on ASR personalization task.
We describe the multi-GPU gradient boosting algorithm implemented in the XGBoost library (https://github.com/dmlc/xgboost). Our algorithm allows fast, scalable training on multi-GPU systems with all of the features of the XGBoost library. We employ data compression techniques to minimise the usage of scarce GPU memory …
Despite showing state-of-the-art performance, deep learning for speech recognition remains challenging to deploy in on-device edge scenarios such as mobile and other consumer devices. Recently, there have been greater efforts in the design of small, low-footprint deep neural networks (DNNs) that are more appropriate fo…
Two new methods reduce communication costs in federated learning.
problem Heavy communication costs in federated learning, especially with non-IID data.
method FedMMD uses MMD constraint for two-stream model training; FedFusion aggregates local and global features.
result FedMMD and FedFusion reduce communication costs by 20% and 60% respectively.
Improves E2E ASR performance on numeric sequences with additional training data and denormalization.
problem Challenges in recognizing numeric sequences out-of-vocabulary in ASR systems.
method Uses text-to-speech for additional numeric training data and a small-footprint neural network for denormalization.
result Reduction of WER by up to a factor of 8 in the longest numeric sequences.
BottleNet++ compresses deep learning features for efficient mobile inference.
problem Balancing computation and communication in mobile devices with deep learning models.
method End-to-end architecture with encoder, non-trainable channel layer, and decoder.
result Achieves high compression ratio and bandwidth reduction with minimal accuracy loss.
On-device machine learning (ML) enables the training process to exploit a massive amount of user-generated private data samples. To enjoy this benefit, inter-device communication overhead should be minimized. With this end, we propose federated distillation (FD), a distributed model training algorithm whose communicati…
Most of the parameters in large vocabulary models are used in embedding layer to map categorical features to vectors and in softmax layer for classification weights. This is a bottle-neck in memory constraint on-device training applications like federated learning and on-device inference applications like automatic spe…
MDLdroid improves mobile deep learning for personal sensing with faster training.
problem Continuous local changes and resource constraints in personal mobile sensing affect global model performance.
method ChainSGD-reduce approach to reduce overhead and balance resources.
result 2x to 3.5x faster training on off-the-shelf mobile devices compared to single-device training.
FedML aims to improve FL research by providing a library and benchmark.
problem Inconsistent FL algorithm development and performance comparison.
method FedML offers an open research library and benchmark supporting diverse computing paradigms and flexible API design.
result FedML facilitates fair algorithm comparison and development in federated learning.
PairNets optimize AI models for fast IoT applications.
problem Slow training and high memory usage of deep neural networks.
method Developed Pairwise Neural Networks (PairNets) with low memory and fast training.
result PairNets achieve faster training (one epoch) and lower prediction errors.
The paper uses quaternions to model quantum learning on devices.
problem Designing adaption and optimization techniques for quantum learning machines.
method Division algebra of quaternions to model computation and measurement on qubits, developing a training framework.
result Established quantum information processing units similar to neurons in classical approaches.
Survey of knowledge distillation for model compression.
problem Deploying large deep models on resource-limited devices.
method Knowledge distillation from large teacher models to small student models.
result Effective model compression and acceleration.
LightSecAgg reduces secure aggregation complexity in FL.
problem Complexity in secure aggregation protocols for FL systems.
method One-shot aggregate-mask reconstruction via mask encoding/decoding.
result Significantly reduces overhead for resiliency against dropped users.
New method FedEx accelerates federated hyperparameter tuning.
problem Federated hyperparameter tuning challenges in distributed learning.
method FedEx method connecting to weight-sharing, adapted for federated optimization.
result FedEx outperforms natural baselines on various benchmarks.
Operating deep neural networks on devices with limited resources requires the reduction of their memory footprints and computational requirements. In this paper we introduce a training method, called look-up table quantization, LUT-Q, which learns a dictionary and assigns each weight to one of the dictionary's values. …
This paper introduces intermittent learning - the goal of which is to enable energy harvested computing platforms capable of executing certain classes of machine learning tasks effectively and efficiently. We identify unique challenges to intermittent learning relating to the data and application semantics of machine l…
Federated learning supports exact support recovery with minimal communication.
problem Learning the exact support of sparse linear regression in federated learning.
method One-shot communication algorithm for exact support recovery without optimization.
result Polynomial sample complexity and logarithmic number of clients required.
Efficient equivariant MobileNetV2 for medical applications on mobile devices.
problem Limited computational resources for deep learning models in medical applications.
method Design and optimize an equivariant version of MobileNetV2 with model quantization.
result Achieved close-to state-of-the-art performance on medical dataset with improved efficiency.
DONNA rapidly finds optimal neural networks across diverse spaces.
problem Efficient scaling and handling of diverse architectural search-spaces in NAS.
method Three-phase pipeline: accuracy predictor, rapid evolutionary search, and optimal model finetuning.
result 100x faster than MNasNet in finding state-of-the-art architectures on-device.
DQA efficiently quantizes deep neural network activations for resource-constrained devices.
problem Efficiently quantizing deep neural network activations for resource-constrained devices.
method DQA uses simple shifting-based operations and Huffman coding for sub-6-bit quantization.
result DQA achieves significantly better accuracy than direct quantization and state-of-the-art methods.
New pruning methods improve energy efficiency of neural networks.
problem Energy-efficient neural networks for devices with limited resources.
method Magnitude and Gradient based pruning at initialization and training of sparse architectures.
result Proposed novel pruning methods prevent full layer pruning and improve training.
Personal VAD detects target speaker voice activity efficiently.
problem Efficiently detect target speaker voice activity for reduced computational cost and battery usage.
method Trains a neural network conditioned on speaker embedding or verification score, outputs probabilities for three speech classes.
result Trained model with 130K parameters outperforms combined standard VAD and speaker recognition networks.
ONLAD Core detects anomalies in edge devices with fast learning and low power.
problem Anomaly detection in edge devices with concept drift and data transfers.
method Highly optimized neural network-based anomaly detection on edge devices.
result ONLAD Core achieves fast anomaly detection and low power consumption.
Recurrent Neural Networks (RNNs) are used in state-of-the-art models in domains such as speech recognition, machine translation, and language modelling. Sparsity is a technique to reduce compute and memory requirements of deep learning models. Sparse RNNs are easier to deploy on devices and high-end server processors. …
Decentralized machine learning is a promising emerging paradigm in view of global challenges of data ownership and privacy. We consider learning of linear classification and regression models, in the setting where the training data is decentralized over many user devices, and the learning algorithm must run on-device, …