Fairness and robustness are two important concerns for federated learning systems. In this work, we identify that robustness to data and model poisoning attacks and fairness, measured as the uniformity of performance across devices, are competing constraints in statistically heterogeneous networks. To address these constraints, we propose employing a simple, general framework for personalized federated learning, Ditto, that can inherently provide fairness and robustness benefits, and develop a scalable solver for it. Theoretically, we analyze the ability of Ditto to achieve fairness and robustness simultaneously on a class of linear problems. Empirically, across a suite of federated datasets, we show that Ditto not only achieves competitive performance relative to recent personalization methods, but also enables more accurate, robust, and fair models relative to state-of-the-art fair or robust baselines.
A wide variety of methods have been developed for identifying depression, but they focus primarily on measuring the degree to which individuals are suffering from depression currently. In this work we explore the possibility of predicting future depression using machine learning applied to longitudinal socio-demographic data. In doing so we show that data such as housing status, and the details of the family environment, can provide cues for predicting future psychiatric disorders. To this end, we introduce a novel deep multi-task recurrent neural network to learn time-dependent depression cues. The depression prediction task is jointly optimized with two auxiliary anomaly ranking tasks, including contrastive one-class feature ranking and deviation ranking. The auxiliary tasks address two key challenges of the problem: 1) the high within class variance of depression samples: they enable the learning of representations that are robust to highly variant in-class distribution of the depression samples; and 2) the small labeled data volume: they significantly enhance the sample efficiency of the prediction model, which reduces the reliance on large depression-labeled datasets that are difficult to collect in practice. Extensive empirical results on large-scale child depression data show that our model is sample-efficient and can accurately predict depression 2-4 years before the illness occurs, substantially outperforming eight representative comparators.
Multi-task learning is widely used in computer vision. Currently, object detection models utilize shared feature map to complete classification and localization tasks simultaneously. By comparing the performance between the original Faster R-CNN and that with partially separated feature maps, we show that: (1) Sharing high-level features for the classification and localization tasks is sub-optimal; (2) Large stride is beneficial for classification but harmful for localization; (3) Global context information could improve the performance of classification. Based on these findings, we proposed a paradigm called Gap-optimized region based convolutional network (G-RCN), which aims to separating these two tasks and optimizing the gap between them. The paradigm was firstly applied to correct the current ResNet protocol by simply reducing the stride and moving the Conv5 block from the head to the feature extraction network, which brings 3.6 improvement of AP70 on the PASCAL VOC dataset and 1.5 improvement of AP on the COCO dataset for ResNet50. Next, the new method is applied on the Faster R-CNN with backbone of VGG16,ResNet50 and ResNet101, which brings above 2.0 improvement of AP70 on the PASCAL VOC dataset and above 1.9 improvement of AP on the COCO dataset. Noticeably, the implementation of G-RCN only involves a few structural modifications, with no extra module added.
In this paper, we present our solution to the Traffic4cast2020 traffic prediction challenge. In this competition, participants are to predict future traffic parameters (speed and volume) in three different cities: Berlin, Istanbul and Moscow. The information provided includes nine channels where the first eight represent the speed and volume for four different direction of traffic (NE, NW, SE and SW), while the last channel is used to indicate presence of traffic incidents. The expected output should have the first 8 channels of the input at six future timing intervals (5, 10, 15, 30, 45, and 60min), while a one hour duration of past traffic data, in 5mins intervals, are provided as input. We solve the problem using a novel sedenion U-Net neural network. Sedenion networks provide the means for efficient encoding of correlated multimodal datasets. We use 12 of the 15 sedenion imaginary parts for the dynamic inputs and the real sedenion component is used for the static input. The sedenion output of the network is used to represent the multimodal traffic predictions. Proposed system achieved a validation MSE of 1.33e-3 and a test MSE of 1.31e-3.
Pedestrians are arguably one of the most safety-critical road users to consider for autonomous vehicles in urban areas. In this paper, we address the problem of jointly detecting pedestrians and recognizing 32 pedestrian attributes from a single image. These encompass visual appearance and behavior, and also include the forecasting of road crossing, which is a main safety concern. For this, we introduce a Multi-Task Learning (MTL) model relying on a composite field framework, which achieves both goals in an efficient way. Each field spatially locates pedestrian instances and aggregates attribute predictions over them. This formulation naturally leverages spatial context, making it well suited to low resolution scenarios such as autonomous driving. By increasing the number of attributes jointly learned, we highlight an issue related to the scales of gradients, which arises in MTL with numerous tasks. We solve it by normalizing the gradients coming from different objective functions when they join at the fork in the network architecture during the backward pass, referred to as fork-normalization. Experimental validation is performed on JAAD, a dataset providing numerous attributes for pedestrian analysis from autonomous vehicles, and shows competitive detection and attribute recognition results, as well as a more stable MTL training.
Zero-shot intent detection (ZSID) aims to deal with the continuously emerging intents without annotated training data. However, existing ZSID systems suffer from two limitations: 1) They are not good at modeling the relationship between seen and unseen intents. 2) They cannot effectively recognize unseen intents under the generalized intent detection (GZSID) setting. A critical problem behind these limitations is that the representations of unseen intents cannot be learned in the training stage. To address this problem, we propose a novel framework that utilizes unseen class labels to learn Class-Transductive Intent Representations (CTIR). Specifically, we allow the model to predict unseen intents during training, with the corresponding label names serving as input utterances. On this basis, we introduce a multi-task learning objective, which encourages the model to learn the distinctions among intents, and a similarity scorer, which estimates the connections among intents more accurately. CTIR is easy to implement and can be integrated with existing methods. Experiments on two real-world datasets show that CTIR brings considerable improvement to the baseline systems.
One crucial challenge of real-world multilingual speech recognition is the long-tailed distribution problem, where some resource-rich languages like English have abundant training data, but a long tail of low-resource languages have varying amounts of limited training data. To overcome the long-tail problem, in this paper, we propose Adapt-and-Adjust (A2), a transformer-based multi-task learning framework for end-to-end multilingual speech recognition. The A2 framework overcomes the long-tail problem via three techniques: (1) exploiting a pretrained multilingual language model (mBERT) to improve the performance of low-resource languages; (2) proposing dual adapters consisting of both language-specific and language-agnostic adaptation with minimal additional parameters; and (3) overcoming the class imbalance, either by imposing class priors in the loss during training or adjusting the logits of the softmax output during inference. Extensive experiments on the CommonVoice corpus show that A2 significantly outperforms conventional approaches.
To cope with reverberation and noise in single channel acoustic scenarios, typical supervised deep neural network~(DNN)-based techniques learn a mapping from reverberant and noisy input features to a user-defined target. Commonly used targets are the desired signal magnitude, a time-frequency mask such as the Wiener gain, or the interference power spectral density and signal-to-interference ratio that can be used to compute a time-frequency mask. In this paper, we propose to incorporate multi-task learning in such DNN-based enhancement techniques by using speech presence probability (SPP) estimation as a secondary task assisting the target estimation in the main task. The advantage of multi-task learning lies in sharing domain-specific information between the two tasks (i.e., target and SPP estimation) and learning more generalizable and robust representations. To simultaneously learn both tasks, we propose to use the adaptive weighting method of losses derived from the homoscedastic uncertainty of tasks. Simulation results show that the dereverberation and noise reduction performance of a single-task DNN trained to directly estimate the Wiener gain is higher than the performance of single-task DNNs trained to estimate the desired signal magnitude, the interference power spectral density, or the signal-to-interference ratio. Incorporating the proposed multi-task learning scheme to jointly estimate the Wiener gain and the SPP increases the dereverberation and noise reduction further.
Deep learning based speech denoising still suffers from the challenge of improving perceptual quality of enhanced signals. We introduce a generalized framework called Perceptual Ensemble Regularization Loss (PERL) built on the idea of perceptual losses. Perceptual loss discourages distortion to certain speech properties and we analyze it using six large-scale pre-trained models: speaker classification, acoustic model, speaker embedding, emotion classification, and two self-supervised speech encoders (PASE+, wav2vec 2.0). We first build a strong baseline (w/o PERL) using Conformer Transformer Networks on the popular enhancement benchmark called VCTK-DEMAND. Using auxiliary models one at a time, we find acoustic event and self-supervised model PASE+ to be most effective. Our best model (PERL-AE) only uses acoustic event model (utilizing AudioSet) to outperform state-of-the-art methods on major perceptual metrics. To explore if denoising can leverage full framework, we use all networks but find that our seven-loss formulation suffers from the challenges of Multi-Task Learning. Finally, we report a critical observation that state-of-the-art Multi-Task weight learning methods cannot outperform hand tuning, perhaps due to challenges of domain mismatch and weak complementarity of losses.
In this paper, we propose two novel and practical deep-learning-based algorithms to solve the wireless channel type (WCT) recognition problem. Specifically, the WCT recognition problem is recast as a classification problem in deep learning due to their similarities, where a deep neural network (DNN) is trained off-line with a diversity of typical WCTs for fifth-generation (5G) and beyond-5G wireless communications, which is then utilized to perform online WCT determination. In the first algorithm, one WCT is regarded as a single task. While in the second scheme, one WCT is jointly characterized by several independent features, each of which is treated as a task and is classified respectively by training a DNN in a multi-task-learning manner, and the final WCT is identified by the combination of those channel features. Simulation results show that the proposed algorithms can classify various WCTs instantaneously with high accuracy, result in satisfactory block error rate and throughput, and outperform a representative baseline WCT determination scheme.
There are two sub-tasks implied in the weakly-supervised SED: audio tagging and event boundary detection. Current methods which combine multi-task learning with SED requires annotations both for these two sub-tasks. Since there are only annotations for audio tagging available in weakly-supervised SED, we design multiple branches with different learning purposes instead of pursuing multiple tasks. Similar to multiple tasks, multiple different learning purposes can also prevent the common feature which the multiple branches share from overfitting to any one of the learning purposes. We design these multiple different learning purposes based on combinations of different MIL strategies and different pooling methods. Experiments on the DCASE 2018 Task 4 dataset and the URBAN-SED dataset both show that our method achieves competitive performance.
Annotation cost is a bottleneck for collecting massive data in mammography, especially for training deep neural networks. In this paper, we study the use of heterogeneous levels of annotation granularity to improve predictive performances. More precisely, we introduce a multi-task learning scheme for training convolutional neural network (ConvNets), which combines segmentation and classification, using image-level and pixel-level annotations. In this way, different objectives can be used to regularize training by sharing intermediate deep representations. Successful experiments are carried out on the Digital Database of Screening Mammography (DDSM) to validate the relevance of the proposed approach.
We propose a learning method well-suited to infer the presence of Tuberculosis (TB) manifestations on Computer Tomography (CT) scans mimicking the radiologist reports. Latent features are extracted from the CT volumes employing the V-Net encoder and those are the input to a Feed-Forward Neural Network (FNN) for multi-class classification. To overtake the issues (e.g., exploding/vanishing gradients, lack of sensibility) that normally appear when training deep 3D models with datasets of limited size and composed of large volumes, our proposal employs: 1) At the network architecture level, the scaled exponential linear unit (SELU) activation which allows the network self-normalization, and 2) at the learning phase, multi-task learning with a loss function weighted by the task homoscedastic uncertainty. The results achieve F1-scores close to or above 0.9 for the detection of TB lesions and a Root Mean Square Error of 1.16 for the number of nodules.
The majority of sound scene analysis work focuses on one of two clearly defined tasks: acoustic scene classification or sound event detection. Whilst this separation of tasks is useful for problem definition, they inherently ignore some subtleties of the real-world, in particular how humans vary in how they describe a scene. Some will describe the weather and features within it, others will use a holistic descriptor like `park', and others still will use unique identifiers such as cities or names. In this paper, we undertake the task of automatic city classification to ask whether we can recognize a city from a set of sound scenes? In this problem each city has recordings from multiple scenes. We test a series of methods for this novel task and show that a simple convolutional neural network (CNN) can achieve accuracy of 50%. This is less than the acoustic scene classification task baseline in the DCASE 2018 ASC challenge on the same data. A simple adaptation to the class labels of pairing city labels with grouped scenes, accuracy increases to 52%, closer to the simpler scene classification task. Finally we also formulate the problem in a multi-task learning framework and achieve an accuracy of 56%, outperforming the aforementioned approaches.
Ultrasound Shear Wave Elastography (USWE) with conventional B-mode imaging demonstrates better performance in lesion segmentation and classification problems. In this article, we propose SHEAR-net, an end-to-end deep neural network, to reconstruct USWE images from tracked tissue displacement data at different time instants induced by a single acoustic radiation force (ARF) with 100% or 50% of the energy in conventional use. The SHEAR-net consists of a localizer called the S-net to first localize the lesion location and then uses recurrent layers to extract temporal correlations from wave patterns using different time frames, and finally, with an estimator, it reconstructs the shear modulus image from the concatenated outputs of S-net and recurrent layers. The network is trained with 800 simulation and a limited number of CIRS tissue mimicking phantom data and is optimized using a multi-task learning loss function where the tasks are: inclusion localization and modulus estimation. The efficacy of the proposed SHEAR-net is extensively evaluated both qualitatively and quantitatively on 125 test set of motion data obtained from simulation and CIRS phantoms. We show that the proposed approach consistently outperforms the current state-of-the-art method and achieves overall 4-5 dB improvement in PSNR and SNR. In addition, an average gain of 0.15 in DSC and SSIM values indicate that the SHEAR-net has a better inclusion coverage area and structural similarity of the two approaches. The proposed real-time deep learning based technique can accurately estimate shear modulus for a minimum tissue displacement of 0.5$\mu$m and image multiple inclusions with a single push ARF.
Multi-modal brain functional connectivity (FC) data have shown great potential for providing insights into individual variations in behavioral and cognitive traits. The joint learning of multi-modal imaging data can utilize the intrinsic association, and thus can boost the learning performance. Although several multi-task based learning models have already been proposed by viewing the feature learning on each modality as one task, most of them ignore the geometric structure information inherent in the modalities, which may play an important role in extracting discriminative features. In this paper, we propose a new manifold regularized multi-task learning model by simultaneously considering between-subject and between-modality relationships. Besides employing a group-sparsity regularizer to jointly select a few common features across multiple tasks (modalities), we design a novel manifold regularizer to preserve the structure information both within and between modalities in our model. This will make our model more adaptive for realistic data analysis. Our model is then validated on the Philadelphia Neurodevelopmental Cohort dataset, where we regard our modalities as functional MRI (fMRI) data collected under two paradigms. Specifically, we conduct experimental studies on fMRI based FC network data in two task conditions for intelligence quotient (IQ) prediction. The results demonstrate that our proposed model can not only achieve improved prediction performance, but also yield a set of IQ-relevant biomarkers.
An intelligent optical performance monitor using multi-task learning based artificial neural network (MTL-ANN) is designed for simultaneous OSNR monitoring and modulation format identification (MFI). Signals' amplitude histograms (AHs) after constant module algorithm are selected as the input features for MTL-ANN. The experimental results of 20-Gbaud NRZ-OOK, PAM4 and PAM8 signals demonstrate that MTL-ANN could achieve OSNR monitoring and MFI simultaneously with higher accuracy and stability compared with single-task learning based ANNs (STL-ANNs). The results show an MFI accuracy of 100% and OSNR monitoring root-mean-square error of 0.63 dB for the three modulation formats under consideration. Furthermore, the number of neuron needed for the single MTL-ANN is almost the half of STL-ANN, which enables reduced-complexity optical performance monitoring devices for real-time performance monitoring.
End-to-end automatic speech recognition (ASR) commonly transcribes audio signals into sequences of characters while its performance is evaluated by measuring the word-error rate (WER). This suggests that predicting sequences of words directly may be helpful instead. However, training with word-level supervision can be more difficult due to the sparsity of examples per label class. In this paper we analyze an end-to-end ASR model that combines a word-and-character representation in a multi-task learning (MTL) framework. We show that it improves on the WER and study how the word-level model can benefit from character-level supervision by analyzing the learned inductive preference bias of each model component empirically. We find that by adding character-level supervision, the MTL model interpolates between recognizing more frequent words (preferred by the word-level model) and shorter words (preferred by the character-level model).
With an aging and growing population, the number of women requiring either screening or symptomatic mammograms is increasing. To reduce the number of mammograms that need to be read by a radiologist while keeping the diagnostic accuracy the same or better than current clinical practice, we develop Man and Machine Mammography Oracle (MAMMO) - a clinical decision support system capable of triaging mammograms into those that can be confidently classified by a machine and those that cannot be, thus requiring the reading of a radiologist. The first component of MAMMO is a novel multi-view convolutional neural network (CNN) with multi-task learning (MTL). MTL enables the CNN to learn the radiological assessments known to be associated with cancer, such as breast density, conspicuity, suspicion, etc., in addition to learning the primary task of cancer diagnosis. We show that MTL has two advantages: 1) learning refined feature representations associated with cancer improves the classification performance of the diagnosis task and 2) issuing radiological assessments provides an additional layer of model interpretability that a radiologist can use to debug and scrutinize the diagnoses provided by the CNN. The second component of MAMMO is a triage network, which takes as input the radiological assessment and diagnostic predictions of the first network's MTL outputs and determines which mammograms can be correctly and confidently diagnosed by the CNN and which mammograms cannot, thus needing to be read by a radiologist. Results obtained on a private dataset of 8,162 patients show that MAMMO reduced the number of radiologist readings by 42.8% while improving the overall diagnostic accuracy in comparison to readings done by radiologists alone. We analyze the triage of patients decided by MAMMO to gain a better understanding of what unique mammogram characteristics require radiologists' expertise.
When performing data classification over a stream of continuously occurring instances, a key challenge is to develop an open-world classifier that anticipates instances from an unknown class. Studies addressing this problem, typically called novel class detection, have considered classification methods that reactively adapt to such changes along the stream. Importantly, they rely on the property of cohesion and separation among instances in feature space. Instances belonging to the same class are assumed to be closer to each other (cohesion) than those belonging to different classes (separation). Unfortunately, this assumption may not have large support when dealing with high dimensional data such as images. In this paper, we address this key challenge by proposing a semisupervised multi-task learning framework called CSIM which aims to intrinsically search for a latent space suitable for detecting labels of instances from both known and unknown classes. Particularly, we utilize a convolution neural network layer that aids in the learning of a latent feature space suitable for novel class detection. We empirically measure the performance of CSIM over multiple realworld image datasets and demonstrate its superiority by comparing its performance with existing semi-supervised methods.