Search

Recently tellurium (Te) has attracted resurgent interests due to its p-type characteristics and outstanding ambient environmental stability. Here we present a substrate engineering based physical vapor deposition method to synthesize high-quality Te nanoflakes and achieved a field-effect hole mobility of 1500 cm2/Vs, which is, to the best of our knowledge, the highest among the existing synthesized van der Waals p-type semiconductors. The high mobility Te enables the fabrication of Te/MoS2 pn diodes with highly gate-tunable electronic and optoelectronic characteristics. The Te/MoS2 heterostructure can be used as a visible range photodetector with a current responsivity up to 630 A/W, which is about one order of magnitude higher than the one achieved using p-type Si-MoS2 PN photodiodes. The photo response of the Te/MoS2 heterojunction also exhibits strong gate tunability due to their ultrathin thickness and unique band structures. The successful synthesis of high mobility Te and the enabled Te/MoS2 photodiodes show promise for the development of highly tunable and ultrathin photodetectors.

A soliton microcomb can play a crucial role in narrow-grid optical communications by replacing many independently operated lasers in wavelength-division multiplexing systems. In this work, we designed and demonstrated power-efficient soliton microcombs with 100-GHz free spectral range in an integrated 4H-SiC platform for the first time. The combination of enabling technologies, including efficient fiber coupling (3 dB insertion loss), high-quality-factor microrings (intrinsic quality factors up to 5.7 million), and the employment of the Raman effect for adiabatic accessing of the soliton state, has enabled the demonstration of soliton pump power as low as 6 mW while supporting comb powers above -20 dBm per line near the pump wavelength.

Deep neural networks often severely forget previously learned knowledge when learning new knowledge. Various continual learning (CL) methods have been proposed to handle such a catastrophic forgetting issue from different perspectives and achieved substantial improvements. In this paper, a novel two-branch continual learning framework is proposed to further enhance most existing CL methods. Specifically, the main branch can be any existing CL model and the newly introduced side branch is a lightweight convolutional network. The output of each main branch block is modulated by the output of the corresponding side branch block. Such a simple two-branch model can then be easily implemented and learned with the vanilla optimization setting without whistles and bells. Extensive experiments with various settings on multiple image datasets show that the proposed framework yields consistent improvements over state-of-the-art methods., Comment: 5 pages,3 figures,accepted by ICASSP 2024

Automatic radiology report generation is booming due to its huge application potential for the healthcare industry. However, existing computer vision and natural language processing approaches to tackle this problem are limited in two aspects. First, when extracting image features, most of them neglect multi-view reasoning in vision and model single-view structure of medical images, such as space-view or channel-view. However, clinicians rely on multi-view imaging information for comprehensive judgment in daily clinical diagnosis. Second, when generating reports, they overlook context reasoning with multi-modal information and focus on pure textual optimization utilizing retrieval-based methods. We aim to address these two issues by proposing a model that better simulates clinicians' perspectives and generates more accurate reports. Given the above limitation in feature extraction, we propose a Globally-intensive Attention (GIA) module in the medical image encoder to simulate and integrate multi-view vision perception. GIA aims to learn three types of vision perception: depth view, space view, and pixel view. On the other hand, to address the above problem in report generation, we explore how to involve multi-modal signals to generate precisely matched reports, i.e., how to integrate previously predicted words with region-aware visual content in next word prediction. Specifically, we design a Visual Knowledge-guided Decoder (VKGD), which can adaptively consider how much the model needs to rely on visual information and previously predicted text to assist next word prediction. Hence, our final Intensive Vision-guided Network (IVGN) framework includes a GIA-guided Visual Encoder and the VKGD. Experiments on two commonly-used datasets IU X-Ray and MIMIC-CXR demonstrate the superior ability of our method compared with other state-of-the-art approaches., Comment: Accepted by Physics in Medicine & Biology

The existing silicon-carbide-on-insulator photonic platform utilizes a thin layer of silicon dioxide under silicon carbide to provide optical confinement and mode isolation. Here, we replace the underneath silicon dioxide layer with a 1-$\mu$m-thick aluminum nitride and demonstrate a 4H-silicon-carbide-on-aluminum-nitride integrated photonics platform for the first time. Efficient grating couplers, low-loss waveguides, and compact microring resonators with intrinsic quality factors up to 210,000 are fabricated. In addition, by undercutting the aluminum nitride layer, the intrinsic quality factor of the silicon carbide microring is improved by nearly one order of magnitude (1.8 million). Finally, an optical pump-probe method is developed to measure the thermal conductivity of the aluminum nitride layer, which is estimated to be over 30 times of that of silicon dioxide.

This research presents a novel active detection model utilizing deep reinforcement learning to accurately detect traffic objects in real-world scenarios. The model employs a deep Q-network based on LSTM-CNN that identifies and aligns target zones with specific categories of traffic objects through implementing a top-down approach with efficient feature extraction of the environment. The model integrates historical and current actions and observations to make a comprehensive analysis. The design of the state space and reward function takes into account the impact of time steps to enable the model to complete the task in fewer steps. Tests conducted demonstrate the model's proficiency, exhibiting exceptional precision and performance in locating traffic signal lights and speed limit signs. The findings of this study highlight the efficacy and potential of the deep reinforcement learning-based active detection model in traffic-related applications, underscoring its robust detection abilities and promising performance., Comment: 14 pages, 4 figures, having passed the preliminary review by experts, about to be submitted to a relevant conference

In this work, we present ODHD, an algorithm for outlier detection based on hyperdimensional computing (HDC), a non-classical learning paradigm. Along with the HDC-based algorithm, we propose IM-ODHD, a computing-in-memory (CiM) implementation based on hardware/software (HW/SW) codesign for improved latency and energy efficiency. The training and testing phases of ODHD may be performed with conventional CPU/GPU hardware or our IM-ODHD, SRAM-based CiM architecture using the proposed HW/SW codesign techniques. We evaluate the performance of ODHD on six datasets from different application domains using three metrics, namely accuracy, F1 score, and ROC-AUC, and compare it with multiple baseline methods such as OCSVM, isolation forest, and autoencoder. The experimental results indicate that ODHD outperforms all the baseline methods in terms of these three metrics on every dataset for both CPU/GPU and CiM implementations. Furthermore, we perform an extensive design space exploration to demonstrate the tradeoff between delay, energy efficiency, and performance of ODHD. We demonstrate that the HW/SW codesign implementation of the outlier detection on IM-ODHD is able to outperform the GPU-based implementation of ODHD by at least 331.5x/889x in terms of training/testing latency (and on average 14.0x/36.9x in terms of training/testing energy consumption.

Implementing stimulated Raman scattering in a low-loss microresonator could lead to Raman lasing. Here, we report the demonstration of an efficient Raman laser with $>50 \%$ power efficiency in an integrated silicon carbide platform for the first time. By fine tuning the free spectral range (FSR) of 43-$\mu$m-radius silicon carbide microresonators, the Stokes resonance corresponding to the dominant Raman shift of $777\ \text{cm}^{-1}$ ($23.3$ THz) is aligned to the center of the Raman gain spectrum, resulting in a low power threshold of $2.5$ mW. The peak Raman gain coefficient is estimated to be ($0.75 \pm 0.15) \ \text{cm}/\text{GW}$ in the 1550 nm band, with an approximate full width at half maximum of ($120 \pm 30$) GHz. In addition, the microresonator is designed to exhibit normal dispersion at the pump wavelength near 1550 nm while possessing anomalous dispersion at the first Stokes near 1760 nm. At high enough input powers, a Kerr microcomb is generated by the Stokes signal acting as the secondary pump, which then mixes with the pump laser through four-wave mixing to attain a wider spectral coverage. Furthermore, cascaded Raman lasing and occurrence of multiple Raman shifts, including $204\ \text{cm}^{-1}$ ($6.1$ THz) and $266\ \text{cm}^{-1}$ ($8.0$ THz) transitions, are also observed. Finally, we show that the Stokes Raman could also help broaden the spectrum in a Kerr microcomb which has anomalous dispersion at the pump wavelength. Our example of a 100-GHz-FSR microcomb has a wavelength span from 1200 nm to 1900 nm with 300 mW on-chip power.

Out-of-distribution (OOD) detection is essential when deploying neural networks in the real world. One main challenge is that neural networks often make overconfident predictions on OOD data. In this study, we propose an effective post-hoc OOD detection method based on a new feature masking strategy and a novel logit smoothing strategy. Feature masking determines the important features at the penultimate layer for each in-distribution (ID) class based on the weights of the ID class in the classifier head and masks the rest features. Logit smoothing computes the cosine similarity between the feature vector of the test sample and the prototype of the predicted ID class at the penultimate layer and uses the similarity as an adaptive temperature factor on the logit to alleviate the network's overconfidence prediction for OOD data. With these strategies, we can reduce feature activation of OOD data and enlarge the gap in OOD score between ID and OOD data. Extensive experiments on multiple standard OOD detection benchmarks demonstrate the effectiveness of our method and its compatibility with existing methods, with new state-of-the-art performance achieved from our method. The source code will be released publicly., Comment: 10 pages, 7 figures

Automated medical image classification is the key component in intelligent diagnosis systems. However, most medical image datasets contain plenty of samples of common diseases and just a handful of rare ones, leading to major class imbalances. Currently, it is an open problem in intelligent diagnosis to effectively learn from imbalanced training data. In this paper, we propose a simple yet effective framework, named \textbf{C}lass \textbf{A}ttention to \textbf{RE}gions of the lesion (CARE), to handle data imbalance issues by embedding attention into the training process of \textbf{C}onvolutional \textbf{N}eural \textbf{N}etworks (CNNs). The proposed attention module helps CNNs attend to lesion regions of rare diseases, therefore helping CNNs to learn their characteristics more effectively. In addition, this attention module works only during the training phase and does not change the architecture of the original network, so it can be directly combined with any existing CNN architecture. The CARE framework needs bounding boxes to represent the lesion regions of rare diseases. To alleviate the need for manual annotation, we further developed variants of CARE by leveraging the traditional saliency methods or a pretrained segmentation model for bounding box generation. Results show that the CARE variants with automated bounding box generation are comparable to the original CARE framework with \textit{manual} bounding box annotations. A series of experiments on an imbalanced skin image dataset and a pneumonia dataset indicates that our method can effectively help the network focus on the lesion regions of rare diseases and remarkably improves the classification performance of rare diseases., Comment: Accepted by Neurocomputing on July 2023. 37 pages

Currently intelligent diagnosis systems lack the ability of continually learning to diagnose new diseases once deployed, under the condition of preserving old disease knowledge. In particular, updating an intelligent diagnosis system with training data of new diseases would cause catastrophic forgetting of old disease knowledge. To address the catastrophic forgetting issue, an Adapter-based Continual Learning framework called ACL is proposed to help effectively learn a set of new diseases at each round (or task) of continual learning, without changing the shared feature extractor. The learnable lightweight task-specific adapter(s) can be flexibly designed (e.g., two convolutional layers) and then added to the pretrained and fixed feature extractor. Together with a specially designed task-specific head which absorbs all previously learned old diseases as a single "out-of-distribution" category, task-specific adapter(s) can help the pretrained feature extractor more effectively extract discriminative features between diseases. In addition, a simple yet effective fine-tuning is applied to collaboratively fine-tune multiple task-specific heads such that outputs from different heads are comparable and consequently the appropriate classifier head can be more accurately selected during model inference. Extensive empirical evaluations on three image datasets demonstrate the superior performance of ACL in continual learning of new diseases. The source code is available at https://github.com/GiantJun/CL_Pytorch., Comment: 10 pages

There is an increasing demand for interpretation of model predictions especially in high-risk applications. Various visualization approaches have been proposed to estimate the part of input which is relevant to a specific model prediction. However, most approaches require model structure and parameter details in order to obtain the visualization results, and in general much effort is required to adapt each approach to multiple types of tasks particularly when model backbone and input format change over tasks. In this study, a simple yet effective visualization framework called PAMI is proposed based on the observation that deep learning models often aggregate features from local regions for model predictions. The basic idea is to mask majority of the input and use the corresponding model output as the relative contribution of the preserved input part to the original model prediction. For each input, since only a set of model outputs are collected and aggregated, PAMI does not require any model detail and can be applied to various prediction tasks with different model backbones and input formats. Extensive experiments on multiple tasks confirm the proposed method performs better than existing visualization approaches in more precisely finding class-specific input regions, and when applied to different model backbones and input formats. The source code will be released publicly., Comment: 19 pages

Current deep learning models often suffer from catastrophic forgetting of old knowledge when continually learning new knowledge. Existing strategies to alleviate this issue often fix the trade-off between keeping old knowledge (stability) and learning new knowledge (plasticity). However, the stability-plasticity trade-off during continual learning may need to be dynamically changed for better model performance. In this paper, we propose two novel ways to adaptively balance model stability and plasticity. The first one is to adaptively integrate multiple levels of old knowledge and transfer it to each block level in the new model. The second one uses prediction uncertainty of old knowledge to naturally tune the importance of learning new knowledge during model training. To our best knowledge, this is the first time to connect model prediction uncertainty and knowledge distillation for continual learning. In addition, this paper applies a modified CutMix particularly to augment the data for old knowledge, further alleviating the catastrophic forgetting issue. Extensive evaluations on the CIFAR100 and the ImageNet datasets confirmed the effectiveness of the proposed method for continual learning.

Graph neural networks (GNNs) have recently emerged as a promising learning paradigm in learning graph-structured data and have demonstrated wide success across various domains such as recommendation systems, social networks, and electronic design automation (EDA). Like other deep learning (DL) methods, GNNs are being deployed in sophisticated modern hardware systems, as well as dedicated accelerators. However, despite the popularity of GNNs and the recent efforts of bringing GNNs to hardware, the fault tolerance and resilience of GNNs have generally been overlooked. Inspired by the inherent algorithmic resilience of DL methods, this paper conducts, for the first time, a large-scale and empirical study of GNN resilience, aiming to understand the relationship between hardware faults and GNN accuracy. By developing a customized fault injection tool on top of PyTorch, we perform extensive fault injection experiments on various GNN models and application datasets. We observe that the error resilience of GNN models varies by orders of magnitude with respect to different models and application datasets. Further, we explore a low-cost error mitigation mechanism for GNN to enhance its resilience. This GNN resilience study aims to open up new directions and opportunities for future GNN accelerator design and architectural optimization.

Silicon carbide (SiC) recently emerged as a promising photonic and quantum material owing to its unique material properties. Here, we carry out an exploratory investigation of the electro-optic effect in high-quality-factor 4H-SiC microresonators. Our findings confirm the existence of the Pockels effect in 4H-SiC for the first time. The extracted Pockels coefficients show certain variations among 4H-SiC wafers from different manufacturers, with the magnitudes of $r_{13}$ and $r_{33}$ estimated to be in the range of (0.3-0.7) pm/V and (0-0.03) pm/V, respectively.

The unique material property of silicon carbide (SiC) and the recent demonstration of low-loss SiC-on-insulator integrated photonics platform have attracted considerable research interests for chip-scale photonic and quantum applications. Here, we carry out a thorough investigation of the Kerr nonlinearity among 4H-SiC wafers from several major wafer manufacturers, and reveal for the first time that their Kerr nonlinear refractive index can be significantly different. By eliminating various measurement errors in the four-wave mixing experiment and improving the theoretical modeling for high-index-contrast waveguides, the best Kerr nonlinear refractive index of 4H-SiC wafers is estimated to be approximately four times of that of stoichiometric silicon nitride in the telecommunication band. In addition, experimental evidence is developed that the Kerr nonlinearity in 4H-SiC wafers can be stronger along the c-axis than that in the orthogonal direction, a feature that was never reported before.

Data imbalance between common and rare diseases during model training often causes intelligent diagnosis systems to have biased predictions towards common diseases. The state-of-the-art approaches apply a two-stage learning framework to alleviate the class-imbalance issue, where the first stage focuses on training of a general feature extractor and the second stage focuses on fine-tuning the classifier head for class rebalancing. However, existing two-stage approaches do not consider the fine-grained property between different diseases, often causing the first stage less effective for medical image classification than for natural image classification tasks. In this study, we propose embedding metric learning into the first stage of the two-stage framework specially to help the feature extractor learn to extract more discriminative feature representations. Extensive experiments mainly on three medical image datasets show that the proposed approach consistently outperforms existing onestage and two-stage approaches, suggesting that metric learning can be used as an effective plug-in component in the two-stage framework for fine-grained class-imbalanced image classification tasks.

Imbalanced training data is a significant challenge for medical image classification. In this study, we propose a novel Progressive Class-Center Triplet (PCCT) framework to alleviate the class imbalance issue particularly for diagnosis of rare diseases, mainly by carefully designing the triplet sampling strategy and the triplet loss formation. Specifically, the PCCT framework includes two successive stages. In the first stage, PCCT trains the diagnosis system via a class-balanced triplet loss to coarsely separate distributions of different classes. In the second stage, the PCCT framework further improves the diagnosis system via a class-center involved triplet loss to cause a more compact distribution for each class. For the class-balanced triplet loss, triplets are sampled equally for each class at each training iteration, thus alleviating the imbalanced data issue. For the class-center involved triplet loss, the positive and negative samples in each triplet are replaced by their corresponding class centers, which enforces data representations of the same class closer to the class center. Furthermore, the class-center involved triplet loss is extended to the pair-wise ranking loss and the quadruplet loss, which demonstrates the generalization of the proposed framework. Extensive experiments support that the PCCT framework works effectively for medical image classification with imbalanced training images. On two skin image datasets and one chest X-ray dataset, the proposed approach respectively obtains the mean F1 score 86.2, 65.2, and 90.66 over all classes and 81.4, 63.87, and 81.92 for rare classes, achieving state-of-the-art performance and outperforming the widely used methods for the class imbalance issue.

Accurate automated analysis of electroencephalography (EEG) would largely help clinicians effectively monitor and diagnose patients with various brain diseases. Compared to supervised learning with labelled disease EEG data which can train a model to analyze specific diseases but would fail to monitor previously unseen statuses, anomaly detection based on only normal EEGs can detect any potential anomaly in new EEGs. Different from existing anomaly detection strategies which do not consider any property of unavailable abnormal data during model development, a task-oriented self-supervised learning approach is proposed here which makes use of available normal EEGs and expert knowledge about abnormal EEGs to train a more effective feature extractor for the subsequent development of anomaly detector. In addition, a specific two branch convolutional neural network with larger kernels is designed as the feature extractor such that it can more easily extract both larger scale and small-scale features which often appear in unavailable abnormal EEGs. The effectively designed and trained feature extractor has shown to be able to extract better feature representations from EEGs for development of anomaly detector based on normal data and future anomaly detection for new EEGs, as demonstrated on three EEG datasets. The code is available at https://github.com/ironing/EEG-AD.

Deep learning has shown its human-level performance in various applications. However, current deep learning models are characterised by catastrophic forgetting of old knowledge when learning new classes. This poses a challenge particularly in intelligent diagnosis systems where initially only training data of a limited number of diseases are available. In this case, updating the intelligent system with data of new diseases would inevitably downgrade its performance on previously learned diseases. Inspired by the process of learning new knowledge in human brains, we propose a Bayesian generative model for continual learning built on a fixed pre-trained feature extractor. In this model, knowledge of each old class can be compactly represented by a collection of statistical distributions, e.g. with Gaussian mixture models, and naturally kept from forgetting in continual learning over time. Unlike existing class-incremental learning methods, the proposed approach is not sensitive to the continual learning process and can be additionally well applied to the data-incremental learning scenario. Experiments on multiple medical and natural image classification tasks showed that the proposed approach outperforms state-of-the-art approaches which even keep some images of old classes during continual learning of new classes.

Ensemble learning is a classical learning method utilizing a group of weak learners to form a strong learner, which aims to increase the accuracy of the model. Recently, brain-inspired hyperdimensional computing (HDC) becomes an emerging computational paradigm that has achieved success in various domains such as human activity recognition, voice recognition, and bio-medical signal classification. HDC mimics the brain cognition and leverages high-dimensional vectors (e.g., 10000 dimensions) with fully distributed holographic representation and (pseudo-)randomness. This paper presents the first effort in exploring ensemble learning in the context of HDC and proposes the first ensemble HDC model referred to as EnHDC. EnHDC uses a majority voting-based mechanism to synergistically integrate the prediction outcomes of multiple base HDC classifiers. To enhance the diversity of base classifiers, we vary the encoding mechanisms, dimensions, and data width settings among base classifiers. By applying EnHDC on a wide range of applications, results show that the EnHDC can achieve on average 3.2\% accuracy improvement over a single HDC classifier. Further, we show that EnHDC with reduced dimensionality, e.g., 1000 dimensions, can achieve similar or even surpass the accuracy of baseline HDC with higher dimensionality, e.g., 10000 dimensions. This leads to a 20\% reduction of storage requirement of HDC model, which is key to enabling HDC on low-power computing platforms.

Continually learning to segment more and more types of image regions is a desired capability for many intelligent systems. However, such continual semantic segmentation suffers from the same catastrophic forgetting issue as in continual classification learning. While multiple knowledge distillation strategies originally for continual classification have been well adapted to continual semantic segmentation, they only consider transferring old knowledge based on the outputs from one or more layers of deep fully convolutional networks. Different from existing solutions, this study proposes to transfer a new type of information relevant to knowledge, i.e. the relationships between elements (Eg. pixels or small local regions) within each image which can capture both within-class and between-class knowledge. The relationship information can be effectively obtained from the self-attention maps in a Transformer-style segmentation model. Considering that pixels belonging to the same class in each image often share similar visual properties, a class-specific region pooling is applied to provide more efficient relationship information for knowledge transfer. Extensive evaluations on multiple public benchmarks support that the proposed self-attention transfer method can further effectively alleviate the catastrophic forgetting issue, and its flexible combination with one or more widely adopted strategies significantly outperforms state-of-the-art solutions., Comment: Published in Pattern Recognition Journal

This paper represents the first effort to explore an automated architecture search for hyperdimensional computing (HDC), a type of brain-inspired neural network. Currently, HDC design is largely carried out in an application-specific ad-hoc manner, which significantly limits its application. Furthermore, the approach leads to inferior accuracy and efficiency, which suggests that HDC cannot perform competitively against deep neural networks. Herein, we present a thorough study to formulate an HDC architecture search space. On top of the search space, we apply reinforcement-learning to automatically explore the HDC architectures. The searched HDC architectures show competitive performance on case studies involving a drug discovery dataset and a language recognition task. On the Clintox dataset, which tries to learn features from developed drugs that passed/failed clinical trials for toxicity reasons, the searched HDC architecture obtains the state-of-the-art ROC-AUC scores, which are 0.80% higher than the manually designed HDC and 9.75% higher than conventional neural networks. Similar results are achieved on the language recognition task, with 1.27% higher performance than conventional methods.

Silicon carbide has recently emerged as a promising photonics material due to its unique properties, including possessing strong second- and third-order nonlinear coefficients and hosting various color centers that can be utilized for a wealth of quantum applications. Here, we report the design and demonstration of octave-spanning microcombs in a 4H-silicon-carbide-on-insulator platform for the first time. Such broadband operation is enabled by optimized nanofabrication achieving >1 million intrinsic quality factors in a 36-$\mu$m-radius microring resonator, and careful dispersion engineering by investigating the dispersion properties of different mode families. For example, for the fundamental transverse-electric mode whose dispersion can be tailored by simply varying the microring waveguide width, we realized a microcomb spectrum covering the wavelength range from 1100 nm to 2400 nm with an on-chip power near 120 mW. While the observed comb state is verified to be chaotic and not soliton, attaining such a large bandwidth is a crucial step towards realizing $f$-2$f$ self-referencing. In addition, we have also observed coherent soliton-crystal state for the fundamental transverse-magnetic mode, which exhibits stronger dispersion than the fundamental transverse-electric mode and hence a narrower bandwidth.

Deep learning models have shown their superior performance in various vision tasks. However, the lack of precisely interpreting kernels in convolutional neural networks (CNNs) is becoming one main obstacle to wide applications of deep learning models in real scenarios. Although existing interpretation methods may find certain visual patterns which are associated with the activation of a specific kernel, those visual patterns may not be specific or comprehensive enough for interpretation of a specific activation of kernel of interest. In this paper, a simple yet effective optimization method is proposed to interpret the activation of any kernel of interest in CNN models. The basic idea is to simultaneously preserve the activation of the specific kernel and suppress the activation of all other kernels at the same layer. In this way, only visual information relevant to the activation of the specific kernel is remained in the input. Consistent visual information from multiple modified inputs would help users understand what kind of features are specifically associated with specific kernel. Comprehensive evaluation shows that the proposed method can help better interpret activation of specific kernels than widely used methods, even when two kernels have very similar activation regions from the same input image.

Successful continual learning of new knowledge would enable intelligent systems to recognize more and more classes of objects. However, current intelligent systems often fail to correctly recognize previously learned classes of objects when updated to learn new classes. It is widely believed that such downgraded performance is solely due to the catastrophic forgetting of previously learned knowledge. In this study, we argue that the class confusion phenomena may also play a role in downgrading the classification performance during continual learning, i.e., the high similarity between new classes and any previously learned classes would also cause the classifier to make mistakes in recognizing these old classes, even if the knowledge of these old classes is not forgotten. To alleviate the class confusion issue, we propose a discriminative distillation strategy to help the classify well learn the discriminative features between confusing classes during continual learning. Experiments on multiple natural image classification tasks support that the proposed distillation strategy, when combined with existing methods, is effective in further improving continual learning., Comment: arXiv admin note: text overlap with arXiv:2104.13614

Continual learning of new knowledge over time is one desirable capability for intelligent systems to recognize more and more classes of objects. Without or with very limited amount of old data stored, an intelligent system often catastrophically forgets previously learned old knowledge when learning new knowledge. Recently, various approaches have been proposed to alleviate the catastrophic forgetting issue. However, old knowledge learned earlier is commonly less preserved than that learned more recently. In order to reduce the forgetting of particularly earlier learned old knowledge and improve the overall continual learning performance, we propose a simple yet effective fusion mechanism by including all the previously learned feature extractors into the intelligent model. In addition, a new feature extractor is included to the model when learning a new set of classes each time, and a feature extractor pruning is also applied to prevent the whole model size from growing rapidly. Experiments on multiple classification tasks show that the proposed approach can effectively reduce the forgetting of old knowledge, achieving state-of-the-art continual learning performance.

Despite tremendous efforts, it is very challenging to generate a robust model to assist in the accurate quantification assessment of COVID-19 on chest CT images. Due to the nature of blurred boundaries, the supervised segmentation methods usually suffer from annotation biases. To support unbiased lesion localisation and to minimise the labeling costs, we propose a data-driven framework supervised by only image-level labels. The framework can explicitly separate potential lesions from original images, with the help of a generative adversarial network and a lesion-specific decoder. Experiments on two COVID-19 datasets demonstrate the effectiveness of the proposed framework and its superior performance to several existing methods., Comment: accepted by ISBI 2021