Over the last decades, Stochastic Gradient Descent (SGD) has been intensively studied by the Machine Learning community. Despite its versatility and excellent performance, the optimization of large models via SGD still is a time-consuming task. To reduce training time, it is common to distribute the training process across multiple devices. Recently, it has been shown that the convergence of asynchronous SGD (ASGD) will always be faster than mini-batch SGD. However, despite these improvements in the theoretical bounds, most ASGD convergence-rate proofs still rely on a centralized parameter server, which is prone to become a bottleneck when scaling out the gradient computations across many distributed processes. In this paper, we present a novel convergence-rate analysis for decentralized and asynchronous SGD (DASGD) which does not require partial synchronization among nodes nor restrictive network topologies. Specifically, we provide a bound of $\mathcal{O}(\sigma\epsilon^{-2}) + \mathcal{O}(QS_{avg}\epsilon^{-3/2}) + \mathcal{O}(S_{avg}\epsilon^{-1})$ for the convergence rate of DASGD, where $S_{avg}$ is the average staleness between models, $Q$ is a constant that bounds the norm of the gradients, and $\epsilon$ is a (small) error that is allowed within the bound. Furthermore, when gradients are not bounded, we prove the convergence rate of DASGD to be $\mathcal{O}(\sigma\epsilon^{-2}) + \mathcal{O}(\sqrt{\hat{S}_{avg}\hat{S}_{max}}\epsilon^{-1})$, with $\hat{S}_{max}$ and $\hat{S}_{avg}$ representing a loose version of the average and maximum staleness, respectively. Our convergence proof holds for a fixed stepsize and any non-convex, homogeneous, and L-smooth objective function. We anticipate that our results will be of high relevance for the adoption of DASGD by a broad community of researchers and developers.
Room layout estimation predicts layouts from a single panorama. It requires datasets with large-scale and diverse room shapes to train the models. However, there are significant imbalances in real-world datasets including the dimensions of layout complexity, camera locations, and variation in scene appearance. These issues considerably influence the model training performance. In this work, we propose the imBalance-Aware Room Layout Estimation (iBARLE) framework to address these issues. iBARLE consists of (1) Appearance Variation Generation (AVG) module, which promotes visual appearance domain generalization, (2) Complex Structure Mix-up (CSMix) module, which enhances generalizability w.r.t. room structure, and (3) a gradient-based layout objective function, which allows more effective accounting for occlusions in complex layouts. All modules are jointly trained and help each other to achieve the best performance. Experiments and ablation studies based on ZInD~\cite{cruz2021zillow} dataset illustrate that iBARLE has state-of-the-art performance compared with other layout estimation baselines.
Atrial fibrillation (AF) is one of the most common arrhythmias with challenging public health implications. Therefore, automatic detection of AF episodes on ECG is one of the essential tasks in biomedical engineering. In this paper, we applied the recently introduced method of compressor-based text classification with gzip algorithm for AF detection (binary classification between heart rhythms). We investigated the normalized compression distance applied to RR-interval and $\Delta$RR-interval sequences ($\Delta$RR-interval is the difference between subsequent RR-intervals). Here, the configuration of the k-nearest neighbour classifier, an optimal window length, and the choice of data types for compression were analyzed. We achieved good classification results while learning on the full MIT-BIH Atrial Fibrillation database, close to the best specialized AF detection algorithms (avg. sensitivity = 97.1\%, avg. specificity = 91.7\%, best sensitivity of 99.8\%, best specificity of 97.6\% with fivefold cross-validation). In addition, we evaluated the classification performance under the few-shot learning setting. Our results suggest that gzip compression-based classification, originally proposed for texts, is suitable for biomedical data and quantized continuous stochastic sequences in general.
Federated learning (FL) has gained significant traction as a privacy-preserving algorithm, but the underlying resemblances of federated learning algorithms like Federated averaging (FedAvg) or Federated SGD (Fed SGD) to ensemble learning algorithms have not been fully explored. The purpose of this paper is to examine the application of FL to object detection as a method to enhance generalizability, and to compare its performance against a centralized training approach for an object detection algorithm. Specifically, we investigate the performance of a YOLOv5 model trained using FL across multiple clients and employ a random sampling strategy without replacement, so each client holds a portion of the same dataset used for centralized training. Our experimental results showcase the superior efficiency of the FL object detector's global model in generating accurate bounding boxes for unseen objects, with the test set being a mixture of objects from two distinct clients not represented in the training dataset. These findings suggest that FL can be viewed from an ensemble algorithm perspective, akin to a synergistic blend of Bagging and Boosting techniques. As a result, FL can be seen not only as a method to enhance privacy, but also as a method to enhance the performance of a machine learning model.
While deep neural networks are being utilized heavily for autonomous driving, they need to be adapted to new unseen environmental conditions for which they were not trained. We focus on a safety critical application of lane detection, and propose a lightweight, fully unsupervised, real-time adaptation approach that only adapts the batch-normalization parameters of the model. We demonstrate that our technique can perform inference, followed by on-device adaptation, under a tight constraint of 30 FPS on Nvidia Jetson Orin. It shows similar accuracy (avg. of 92.19%) as a state-of-the-art semi-supervised adaptation algorithm but which does not support real-time adaptation.
Large and sparse feed-forward layers (S-FFN) such as Mixture-of-Experts (MoE) have proven effective in scaling up Transformers model size for \textit{pretraining} large language models. By only activating part of the FFN parameters conditioning on input, S-FFN improves generalization performance while keeping training and inference costs (in FLOPs) fixed. In this work, we analyzed two major design choices of S-FFN: the memory block (a.k.a. expert) size and the memory block selection method under a general conceptual framework of sparse neural memory. Using this unified framework, we compare several S-FFN architectures for language modeling and provide insights into their relative efficacy and efficiency. We found a simpler selection method -- \textbf{\texttt{Avg-K}} that selects blocks through their mean aggregated hidden states, achieving lower perplexity in language model pretraining compared to existing MoE architectures including Switch Transformer (Fedus et al., 2021) and HashLayer (Roller et al., 2021).
We propose a novel hierarchical Bayesian approach to Federated Learning (FL), where our model reasonably describes the generative process of clients' local data via hierarchical Bayesian modeling: constituting random variables of local models for clients that are governed by a higher-level global variate. Interestingly, the variational inference in our Bayesian model leads to an optimisation problem whose block-coordinate descent solution becomes a distributed algorithm that is separable over clients and allows them not to reveal their own private data at all, thus fully compatible with FL. We also highlight that our block-coordinate algorithm has particular forms that subsume the well-known FL algorithms including Fed-Avg and Fed-Prox as special cases. Beyond introducing novel modeling and derivations, we also offer convergence analysis showing that our block-coordinate FL algorithm converges to an (local) optimum of the objective at the rate of $O(1/\sqrt{t})$, the same rate as regular (centralised) SGD, as well as the generalisation error analysis where we prove that the test error of our model on unseen data is guaranteed to vanish as we increase the training data size, thus asymptotically optimal.
Salient Span Masking (SSM) has shown itself to be an effective strategy to improve closed-book question answering performance. SSM extends general masked language model pretraining by creating additional unsupervised training sentences that mask a single entity or date span, thus oversampling factual information. Despite the success of this paradigm, the span types and sampling strategies are relatively arbitrary and not widely studied for other tasks. Thus, we investigate SSM from the perspective of temporal tasks, where learning a good representation of various temporal expressions is important. To that end, we introduce Temporal Span Masking (TSM) intermediate training. First, we find that SSM alone improves the downstream performance on three temporal tasks by an avg. +5.8 points. Further, we are able to achieve additional improvements (avg. +0.29 points) by adding the TSM task. These comprise the new best reported results on the targeted tasks. Our analysis suggests that the effectiveness of SSM stems from the sentences chosen in the training data rather than the mask choice: sentences with entities frequently also contain temporal expressions. Nonetheless, the additional targeted spans of TSM can still improve performance, especially in a zero-shot context.
To investigate whether a deep learning model can detect Covid-19 from disruptions in the human body's physiological (heart rate) and rest-activity rhythms (rhythmic dysregulation) caused by the SARS-CoV-2 virus. We propose CovidRhythm, a novel Gated Recurrent Unit (GRU) Network with Multi-Head Self-Attention (MHSA) that combines sensor and rhythmic features extracted from heart rate and activity (steps) data gathered passively using consumer-grade smart wearable to predict Covid-19. A total of 39 features were extracted (standard deviation, mean, min/max/avg length of sedentary and active bouts) from wearable sensor data. Biobehavioral rhythms were modeled using nine parameters (mesor, amplitude, acrophase, and intra-daily variability). These features were then input to CovidRhythm for predicting Covid-19 in the incubation phase (one day before biological symptoms manifest). A combination of sensor and biobehavioral rhythm features achieved the highest AUC-ROC of 0.79 [Sensitivity = 0.69, Specificity=0.89, F$_{0.1}$ = 0.76], outperforming prior approaches in discriminating Covid-positive patients from healthy controls using 24 hours of historical wearable physiological. Rhythmic features were the most predictive of Covid-19 infection when utilized either alone or in conjunction with sensor features. Sensor features predicted healthy subjects best. Circadian rest-activity rhythms that combine 24h activity and sleep information were the most disrupted. CovidRhythm demonstrates that biobehavioral rhythms derived from consumer-grade wearable data can facilitate timely Covid-19 detection. To the best of our knowledge, our work is the first to detect Covid-19 using deep learning and biobehavioral rhythms features derived from consumer-grade wearable data.
In this paper, we present the Logically submissions to De-Factify 2 challenge (DE-FACTIFY 2023) on the task 1 of Multi-Modal Fact Checking. We describes our submissions to this challenge including explored evidence retrieval and selection techniques, pre-trained cross-modal and unimodal models, and a cross-modal veracity model based on the well established Transformer Encoder (TE) architecture which is heavily relies on the concept of self-attention. Exploratory analysis is also conducted on this Factify 2 data set that uncovers the salient multi-modal patterns and hypothesis motivating the architecture proposed in this work. A series of preliminary experiments were done to investigate and benchmarking different pre-trained embedding models, evidence retrieval settings and thresholds. The final system, a standard two-stage evidence based veracity detection system, yields weighted avg. 0.79 on both val set and final blind test set on the task 1, which achieves 3rd place with a small margin to the top performing system on the leaderboard among 9 participants.
The efficient extraction of text information from the background in degraded color document images is an important challenge in the preservation of ancient manuscripts. The imperfect preservation of ancient manuscripts has led to different types of degradation over time, such as page yellowing, staining, and ink bleeding, seriously affecting the results of document image binarization. This work proposes an effective three-stage network method to image enhancement and binarization of degraded documents using generative adversarial networks (GANs). Specifically, in Stage-1, we first split the input images into multiple patches, and then split these patches into four single-channel patch images (gray, red, green, and blue). Then, three single-channel patch images (red, green, and blue) are processed by the discrete wavelet transform (DWT) with normalization. In Stage-2, we use four independent generators to separately train GAN models based on the four channels on the processed patch images to extract color foreground information. Finally, in Stage-3, we train two independent GAN models on the outputs of Stage-2 and the resized original input images (512x512) as the local and global predictions to obtain the final outputs. The experimental results show that the Avg-Score metrics of the proposed method are 77.64, 77.95, 79.05, 76.38, 75.34, and 77.00 on the (H)-DIBCO 2011, 2013, 2014, 2016, 2017, and 2018 datasets, which are at the state-of-the-art level. The implementation code for this work is available at https://github.com/abcpp12383/ThreeStageBinarization.
The bottleneck of distributed edge learning (DEL) over wireless has shifted from computing to communication, primarily the aggregation-averaging (Agg-Avg) process of DEL. The existing transmission control protocol (TCP)-based data networking schemes for DEL are application-agnostic and fail to deliver adjustments according to application layer requirements. As a result, they introduce massive excess time and undesired issues such as unfairness and stragglers. Other prior mitigation solutions have significant limitations as they balance data flow rates from workers across paths but often incur imbalanced backlogs when the paths exhibit variance, causing stragglers. To facilitate a more productive DEL, we develop a hybrid multipath TCP (MPTCP) by combining model-based and deep reinforcement learning (DRL) based MPTCP for DEL that strives to realize quicker iteration of DEL and better fairness (by ameliorating stragglers). Hybrid MPTCP essentially integrates two radical TCP developments: i) successful existing model-based MPTCP control strategies and ii) advanced emerging DRL-based techniques, and introduces a novel hybrid MPTCP data transport for easing the communication of the Agg-Avg process. Extensive emulation results demonstrate that the proposed hybrid MPTCP can overcome excess time consumption and ameliorate the application layer unfairness of DEL effectively without injecting additional inconstancy and stragglers.
The use of multilingual language models for tasks in low and high-resource languages has been a success story in deep learning. In recent times, Arabic has been receiving widespread attention on account of its dialectal variance. While prior research studies have tried to adapt these multilingual models for dialectal variants of Arabic, it still remains a challenging problem owing to the lack of sufficient monolingual dialectal data and parallel translation data of such dialectal variants. It remains an open problem on whether the limited dialectical data can be used to improve the models trained in Arabic on its dialectal variants. First, we show that multilingual-BERT (mBERT) incrementally pretrained on Arabic monolingual data takes less training time and yields comparable accuracy when compared to our custom monolingual Arabic model and beat existing models (by an avg metric of +$6.41$). We then explore two continual pre-training methods-- (1) using small amounts of dialectical data for continual finetuning and (2) parallel Arabic to English data and a Translation Language Modeling loss function. We show that both approaches help improve performance on dialectal classification tasks ($+4.64$ avg. gain) when used on monolingual models.
Detecting out-of-distribution (OOD) instances is significant for the safe deployment of NLP models. Among recent textual OOD detection works based on pretrained language models (PLMs), distance-based methods have shown superior performance. However, they estimate sample distance scores in the last-layer CLS embedding space and thus do not make full use of linguistic information underlying in PLMs. To address the issue, we propose to boost OOD detection by deriving more holistic sentence embeddings. On the basis of the observations that token averaging and layer combination contribute to improving OOD detection, we propose a simple embedding approach named Avg-Avg, which averages all token representations from each intermediate layer as the sentence embedding and significantly surpasses the state-of-the-art on a comprehensive suite of benchmarks by a 9.33% FAR95 margin. Furthermore, our analysis demonstrates that it indeed helps preserve general linguistic knowledge in fine-tuned PLMs and substantially benefits detecting background shifts. The simple yet effective embedding method can be applied to fine-tuned PLMs with negligible extra costs, providing a free gain in OOD detection. Our code is available at https://github.com/lancopku/Avg-Avg.
Temporal action localization plays an important role in video analysis, which aims to localize and classify actions in untrimmed videos. The previous methods often predict actions on a feature space of a single-temporal scale. However, the temporal features of a low-level scale lack enough semantics for action classification while a high-level scale cannot provide rich details of the action boundaries. To address this issue, we propose to predict actions on a feature space of multi-temporal scales. Specifically, we use refined feature pyramids of different scales to pass semantics from high-level scales to low-level scales. Besides, to establish the long temporal scale of the entire video, we use a spatial-temporal transformer encoder to capture the long-range dependencies of video frames. Then the refined features with long-range dependencies are fed into a classifier for the coarse action prediction. Finally, to further improve the prediction accuracy, we propose to use a frame-level self attention module to refine the classification and boundaries of each action instance. Extensive experiments show that the proposed method can outperform state-of-the-art approaches on the THUMOS14 dataset and achieves comparable performance on the ActivityNet1.3 dataset. Compared with A2Net (TIP20, Avg\{0.3:0.7\}), Sub-Action (CSVT2022, Avg\{0.1:0.5\}), and AFSD (CVPR21, Avg\{0.3:0.7\}) on the THUMOS14 dataset, the proposed method can achieve improvements of 12.6\%, 17.4\% and 2.2\%, respectively
Machine learning has been applied to almost all fields of computer science over the past decades. The introduction of GANs allowed for new possibilities in fields of medical research and text prediction. However, these new fields work with ever more privacy-sensitive data. In order to maintain user privacy, a combination of federated learning, differential privacy and GANs can be used to work with private data without giving away a users' privacy. Recently, two implementations of such combinations have been published: DP-Fed-Avg GAN and GS-WGAN. This paper compares their performance and introduces an alternative version of DP-Fed-Avg GAN that makes use of denoising techniques to combat the loss in accuracy that generally occurs when applying differential privacy and federated learning to GANs. We also compare the novel adaptation of denoised DP-Fed-Avg GAN to the state-of-the-art implementations in this field.
We study the asynchronous stochastic gradient descent algorithm for distributed training over $n$ workers which have varying computation and communication frequency over time. In this algorithm, workers compute stochastic gradients in parallel at their own pace and return those to the server without any synchronization. Existing convergence rates of this algorithm for non-convex smooth objectives depend on the maximum gradient delay $\tau_{\max}$ and show that an $\epsilon$-stationary point is reached after $\mathcal{O}\!\left(\sigma^2\epsilon^{-2}+ \tau_{\max}\epsilon^{-1}\right)$ iterations, where $\sigma$ denotes the variance of stochastic gradients. In this work (i) we obtain a tighter convergence rate of $\mathcal{O}\!\left(\sigma^2\epsilon^{-2}+ \sqrt{\tau_{\max}\tau_{avg}}\epsilon^{-1}\right)$ without any change in the algorithm where $\tau_{avg}$ is the average delay, which can be significantly smaller than $\tau_{\max}$. We also provide (ii) a simple delay-adaptive learning rate scheme, under which asynchronous SGD achieves a convergence rate of $\mathcal{O}\!\left(\sigma^2\epsilon^{-2}+ \tau_{avg}\epsilon^{-1}\right)$, and does not require any extra hyperparameter tuning nor extra communications. Our result allows to show for the first time that asynchronous SGD is always faster than mini-batch SGD. In addition, (iii) we consider the case of heterogeneous functions motivated by federated learning applications and improve the convergence rate by proving a weaker dependence on the maximum delay compared to prior works. In particular, we show that the heterogeneity term in convergence rate is only affected by the average delay within each worker.
This paper investigates different pretraining approaches to spoken language identification. The paper is based on our submission to the Oriental Language Recognition 2021 Challenge. We participated in two tracks of the challenge: constrained and unconstrained language recognition. For the constrained track, we first trained a Conformer-based encoder-decoder model for multilingual automatic speech recognition (ASR), using the provided training data that had transcripts available. The shared encoder of the multilingual ASR model was then finetuned for the language identification task. For the unconstrained task, we relied on both externally available pretrained models as well as external data: the multilingual XLSR-53 wav2vec2.0 model was finetuned on the VoxLingua107 corpus for the language recognition task, and finally finetuned on the provided target language training data, augmented with CommonVoice data. Our primary metric $C_{\rm avg}$ values on the Test set are 0.0079 for the constrained task and 0.0119 for the unconstrained task which resulted in the second place in both rankings. In post-evaluation experiments, we study the amount of target language data needed for training an accurate backend model, the importance of multilingual pretraining data, and compare different models as finetuning starting points.
Generic Boundary Detection (GBD) aims at locating the general boundaries that divide videos into semantically coherent and taxonomy-free units, and could serve as an important pre-processing step for long-form video understanding. Previous works often separately handle these different types of generic boundaries with specific designs of deep networks from simple CNN to LSTM. Instead, in this paper, we present Temporal Perceiver, a general architecture with Transformer, offering a unified solution to the detection of arbitrary generic boundaries, ranging from shot-level, event-level, to scene-level GBDs. The core design is to introduce a small set of latent feature queries as anchors to compress the redundant video input into a fixed dimension via cross-attention blocks. Thanks to this fixed number of latent units, it greatly reduces the quadratic complexity of attention operation to a linear form of input frames. Specifically, to explicitly leverage the temporal structure of videos, we construct two types of latent feature queries: boundary queries and context queries, which handle the semantic incoherence and coherence accordingly. Moreover, to guide the learning of latent feature queries, we propose an alignment loss on the cross-attention maps to explicitly encourage the boundary queries to attend on the top boundary candidates. Finally, we present a sparse detection head on the compressed representation, and directly output the final boundary detection results without any post-processing module. We test our Temporal Perceiver on a variety of GBD benchmarks. Our method obtains the state-of-the-art results on all benchmarks with RGB single-stream features: SoccerNet-v2 (81.9% avg-mAP), Kinetics-GEBD (86.0% avg-f1), TAPOS (73.2% avg-f1), MovieScenes (51.9% AP and 53.1% Miou) and MovieNet (53.3% AP and 53.2% Miou), demonstrating the generalization ability of our Temporal Perceiver.
Measuring rotation speed is essential to many engineering applications; it elicits faults undetectable by vibration monitoring alone and enhances the vibration signal analysis of rotating machines. Optical, magnetic or mechanical Tachometers are currently state-of-art. Their limitations are they require line-of-sight, direct access to the rotating object. This paper proposes RFTacho, a rotation speed measurement \emph{system} that leverages novel hardware and signal processing algorithms to produce highly accurate readings conveniently. RFTacho uses RF Orbital Angular Momentum (OAM) waves to measure rotation speed of multiple machines simultaneously with no requirements from the machine's properties. OAM antennas allow it to operate in high-scattering environments, commonly found in industries, as they are resilient to de-polarization compared to linearly polarized antennas. RFTacho achieves this by using two novel signal processing algorithms to extract the rotation speed of several rotating objects simultaneously amidst noise arising from high-scattering environments, non-line-of-sight scenarios and dynamic environmental conditions with a resolution of $1 rpm$. We test RFTacho on several real-world machines like fans, motors, air conditioners. Results show that RFTacho has avg. error of $<0.5\%$ compared to ground truth. We demonstrate RFTacho's simultaneous multiple-object measurement capability that other tachometers do not have. Initial experiments show that RFTacho can measure speeds as high as 7000 rpm (theoretically 60000 rpm) with high resiliency at different coverage distances and orientation angles, requiring only 150 mW transmit power while operating in the 5 GHz license-exempt band. RFTacho is the first RF-based sensing system that combines OAM waves and novel processing approaches to measure the rotation speed of multiple machines simultaneously in a non-intrusive way.