Medical image segmentation serves as a cornerstone for quantitative analysis in clinical diagnosis, with multi-target joint segmentation playing a critical role in comprehensive disease staging and treatment planning. Existing multi-task learning (MTL) methods primarily focus on balancing competing objectives but often overlook the intrinsic coupling between task-specific optimization and generalization. To address this gap, we propose a novel framework based on worst-case primal-dual optimization, reformulating multi-task coordination as a constrained minimax problem. By introducing adaptive Lagrange multipliers and task-specific parameter updates, our framework ensures balanced task performance and generalizationability across diverse segmentation targets. We develop a family of theoretical and heuristic approaches leveraging worst-case performance to enhance task balancing. Our methods are validated on prostate tumor segmentation and multi-organ segmentation tasks, using both our independently collected ProstCancerPET/CT dataset and the public AutoPETIII challenge dataset. Comprehensive experiments demonstrate that our methods outperform state-of-the-art approaches. Code is available at: https://github.com/p71-23/LoW-Family.

Texture roughness perception is crucial for autonomous robots to perform manipulation, quality inspection, and material discrimination in unknown environments. This work proposes an approach to combine vibration and force data using the VibroTact sensor for texture roughness classification. Vibration and force data are first processed by CNN and ANN models, and then combined using a Bayesian framework. This approach is evaluated by recognizing 15 textures with different roughness (7 soft and 8 hard textures) using individual ANN and CNN models, and is compared against the Bayesian combination of both methods. Texture data is collected by mounting the VibroTact sensor on a robotic arm and using three sliding exploratory procedures (vertical, diagonal, and circular sliding). The texture roughness recognition results achieve 100% accuracy using the combined approach, which improves the performance of individual ANN and CNN models which range from 87.50% to 100% accuracy. The results also show that diagonal and vertical sliding are optimal for recognizing hard and soft textures, respectively. This approach demonstrates its potential for industrial robotics applications that require texture discrimination.

The encoded large DNA can be cloned and stored in vivo, capable of write-once and stable replication for multiple retrievals, offering potential in economic data archiving. Nanopore sequencing is advantageous in data access of large DNA due to its rapidity and long-read sequencing capability. However, the data readout is commonly limited by insertion and deletion (indel) errors and sequence assembly complexity. Here, a pragmatic soft-decision data readout is presented, achieving assembly-free sequence reconstruction, indel error correction, and ultra-low coverage data readout. Specifically, the watermark is cleverly embedded within large DNA fragments, allowing for the direct localization of raw reads via watermark alignment to avoid complex read assembly. A soft-decision forward-backward algorithm is proposed, which can identify indel errors and provide probability information to the error correction code, enabling error-free data recovery. Additionally, a minimum state transition is maintained, and a read segmentation is incorporated to achieve fast information reading. The readout assays for two circular plasmids (~51 kb) with different coding rates were demonstrated and achieved error-free recovery directly from noisy reads (error rate ~1%) at coverage of 1-4×. Simulations conducted on large-scale datasets across various error rates further confirm the scalability of the method and its robust performance under extreme conditions. This readout method enables nearly single-molecule recovery of large DNA, particularly suitable for rapid readout of DNA storage.

Deep learning excels at learning task information from large amounts of data, but struggles with learning from declarative high-level knowledge that can be more succinctly expressed directly. In this work, we introduce PYLON, a neuro-symbolic training framework that builds on PyTorch to augment procedurally trained models with declaratively specified knowledge. PYLON lets users programmatically specify constraints as Python functions and compiles them into a differentiable loss, thus training predictive models that fit the data whilst satisfying the specified constraints. PYLON includes both exact as well as approximate compilers to efficiently compute the loss, employing fuzzy logic, sampling methods, and circuits, ensuring scalability even to complex models and constraints. Crucially, a guiding principle in designing PYLON is the ease with which any existing deep learning codebase can be extended to learn from constraints in a few lines code: a function that expresses the constraint, and a single line to compile it into a loss. Our demo comprises of models in NLP, computer vision, logical games, and knowledge graphs that can be interactively trained using constraints as supervision.

Mesoscale eddy is a typical mesoscale ocean phenomenon, which widely exists in all oceans and marginal seas around the world. The spatiotemporal scale of mesoscale eddies ranges from a few days to hundreds of days, tens of kilometers to hundreds of kilometers. The traditional mesoscale eddy identification is subjective and usually depends on expert discrimination or threshold setting. In this study, due to the significant advantages of YOLO series target recognition models in the field of deep learning, we propose an ocean mesoscale eddy identification algorithm for deep transfer learning target recognition based on YOLOF (You Only Look One Level Feature). Compared with traditional recognition methods, this model has better recognition effect, The influence of setting threshold on mesoscale eddy identification is avoided, and the identification speed is improved to a certain extent.

Question answering systems have become prominent in all areas, while in the medical domain it has been challenging because of the abundant domain knowledge. Retrieval based approach has become promising as large pretrained language models come forth. This study focuses on building a retrieval-based medical question answering system, tackling the challenge with large language models and knowledge extensions via graphs. We first retrieve an extensive but coarse set of answers via Elasticsearch efficiently. Then, we utilize semantic matching with pretrained language models to achieve a fine-grained ranking enhanced with named entity recognition and knowledge graphs to exploit the relation of the entities in question and answer. A new architecture based on siamese structures for answer selection is proposed. To evaluate the approach, we train and test the model on two Chinese data sets, NLPCC2017 and cMedQA. We also conduct experiments on two English data sets, TREC-QA and WikiQA. Our model achieves consistent improvement as compared to strong baselines on all data sets. Qualification studies with cMedQA and our in-house data set show that our system gains highly competitive performance. The proposed medical question answering system outperforms baseline models and systems in quantification and qualification evaluations.

Water quality parameter is a key index to indicate the quality of water and the variation trend of materials in water. One of the parameters, biochemical oxygen demand () is an important parameter reflecting the condition of organic pollution in water. Current inversion methods for water quality parameter mainly include empirical, semi-empirical and physical method. In this study, due to the advantages of hyperspectral remote sensing with a large number of bands and high spectral resolution, we use semi-empirical method combined with hyperspectral remote sensing image to build ratio linear regression model and ratio quadratic polynomial regression model to invert the BOD5 in Qinghai Lake. The comparative analysis shows that the performance of linear model is better than the quadratic polynomial one. The results show that the R2of inversion accuracy is approximately 0.58 and Root Mean Square Error (RMSE) is below 0.16(O2, mg/L), and the inversion results are basically consistent with the spatial distribution of in situ measured BOD5 values in Qinghai Lake.

Qinghai Lake is the largest lake in China, and its water quality has a great impact on the ecological environment of the surrounding area. Chemical oxygen demand (COD) is an important indicator for water quality. In this study, a remote sensing inverse model for COD was established by using the high-spectral images of Zhuhai-1 satellite with measured sample data and a thematic map of the spatial distribution for COD was drawn in Qinghai Lake. The results showed that COD has the greatest correlation with B2 (460nm) and B14 (670nm) band ratio, and the linear regression model is best in all regression models (R2=0.8453, RMSE=1.06 mg/L), and the model results are basically consistent with the actual situation. This work showed that it is feasible to use hyperspectral images to inverse water quality parameters, and monitor the COD spatial distribution and movement of water bodies in a wide and real-time way.

Net Primary Production (NPP) is a key component of the terrestrial carbon cycle. In this study, the CASA model was applied to analyze the NPP change in the Qinghai Lake Basin in 2020, using the meteorological data and GaoFen-1 data. The results show that the NPP value changed significantly through the whole year. It has the highest NPP$(107.6 gC. m^{-2}\cdot month^{-1})$in July, while the NPP value of January is the lowest at 3.6$gC\cdot m^{-2}\cdot month^{-1}$. Spatially, the annual average NPP in Gangcha County is the highest, followed by Haiyan County and Gonghe County, and the value of Tianjun County is the lowest. This study provides an effective indicator for the environmental evaluation of the Qinghai Lake Basin.

This work deals with the challenge of learning and reasoning over language and vision data for the related downstream tasks such as visual question answering (VQA) and natural language for visual reasoning (NLVR).We design a novel cross-modality relevance module that is used in an end-to-end framework to learn the relevance representation between components of various input modalities under the supervision of a target task, which is more generalizable to unobserved data compared to merely reshaping the original representation space.In addition to modeling the relevance between the textual entities and visual entities, we model the higher-order relevance between entity relations in the text and object relations in the image.Our proposed approach shows competitive performance on two different language and vision tasks using public benchmarks and improves the state-of-the-art published results.The learned alignments of input spaces and their relevance representations by NLVR task boost the training efficiency of VQA task.

Structured learning algorithms usually involve an inference phase that selects the best global output variables assignments based on the local scores of all possible assignments. We extend deep neural networks with structured learning to combine the power of learning representations and leveraging the use of domain knowledge in the form of output constraints during training. Introducing a non-differentiable inference module to gradient-based training is a critical challenge. Compared to using conventional loss functions that penalize every local error independently, we propose an inference-masked loss that takes into account the effect of inference and does not penalize the local errors that can be corrected by the inference. We empirically show the inference-masked loss combined with the negative log-likelihood loss improves the performance on different tasks, namely entity relation recognition on CoNLL04 and ACE2005 corpora, and spatial role labeling on CLEF 2017 mSpRL dataset. We show the proposed approach helps to achieve better generalizability, particularly in the low-data regime.

Recent studies in sequence-to-sequence learning demonstrate that RNN encoder-decoder structure can successfully generate Chinese poetry. However, existing methods can only generate poetry with a given first line or user's intent theme. In this paper, we proposed a three-stage multi-modal Chinese poetry generation approach. Given a picture, the first line, the title and the other lines of the poem are successively generated in three stages. According to the characteristics of Chinese poems, we propose a hierarchy-attention seq2seq model which can effectively capture character, phrase, and sentence information between contexts and improve the symmetry delivered in poems. In addition, the Latent Dirichlet allocation (LDA) model is utilized for title generation and improve the relevance of the whole poem and the title. Compared with strong baseline, the experimental results demonstrate the effectiveness of our approach, using machine evaluations as well as human judgments.

Memory is one of the most important mechanisms in recurrent neural networks (RNNs) learning. It plays a crucial role in practical applications, such as sequence learning. With a good memory mechanism, long term history can be fused with current information, and can thus improve RNNs learning. Developing a suitable memory mechanism is always desirable in the field of RNNs. This paper proposes a novel memory mechanism for RNNs. The main contributions of this paper are: 1) an auxiliary memory unit (AMU) is proposed, which results in a new special RNN model (AMU-RNN), separating the memory and output explicitly and 2) an efficient learning algorithm is developed by employing the technique of error flow truncation. The proposed AMU-RNN model, together with the developed learning algorithm, can learn and maintain stable memory over a long time range. This method overcomes both the learning conflict problem and gradient vanishing problem. Unlike the traditional method, which mixes the memory and output with a single neuron in a recurrent unit, the AMU provides an auxiliary memory neuron to maintain memory in particular. By separating the memory and output in a recurrent unit, the problem of learning conflicts can be eliminated easily. Moreover, by using the technique of error flow truncation, each auxiliary memory neuron ensures constant error flow during the learning process. The experiments demonstrate good performance of the proposed AMU-RNNs and the developed learning algorithm. The method exhibits quite efficient learning performance with stable convergence in the AMU-RNN learning and outperforms the state-of-the-art RNN models in sequence generation and sequence classification tasks.

Residual classifiers are common in dictionary-based multiclass classification. This paper proposes the concept of performance functions for residual classifiers. A performance function for multiclass classifications is a conceptual measurement function that combines local and global measurements. In general, the performance function is nonlinear. To explore the properties of the performance function, we employ the Taylor series expansion technique and derive a family of measurement functions. Specifically, the linear measurement and the quadratic measurement (QM) are derived. By exploiting the effect of the higher order terms in the performance function as well as the fundamental nondecreasing constrain, we derive the normalized QM (NQM). We present the classifier for multiclass classification using the proposed measurements. The proposed algorithms are tested against frontal faces and handwritten digit recognition tasks. Our tests show that the QM classifier achieves competitive classification results compared with baseline methods. NQM shows better stability with different parameter configurations.

This paper studies an extension and improvement of the joint detection-decoding algorithm for nonbinary LDPC-coded modulation systems. The iterative joint detection-decoding (IJDD) algorithm in [1] combines nonbinary LDPC decoding with signal detection based on the hard-message passing strategy, resulting in significantly reduced decoding complexity. However, it applies only to majority-logic decodable nonbinary LDPC codes with high column weight. For nonbinary LDPC codes with low column weight, a noticeable performance loss will be incurred. To handle this problem, we propose a reliability-based iterative joint detection-decoding (also termed improved IJDD) algorithm, which combines the accumulated reliability of symbols based on the one-step majority-logic decoding (MLGD) algorithm and a Chase-like local list decoding algorithm. Simulation results show that the improved IJDD algorithm outperforms the IJDD algorithm by about 0.3 dB using nonbinary LDPC codes with high column weight, and by about 3 dB using nonbinary LDPC codes with low column weight (dv= 4), while maintaining the low complexity of decoding. Compared to the FFT-QSPA, the proposed algorithm has a performance degradation of 0.5 dB in the high column weight regime, and about 1 dB in the low column weight regime.

Long-term stress may lead to many severe physical and mental problems. Traditional psychological stress detection usually relies on the active individual participation, which makes the detection labor-consuming, time-costing and hysteretic. With the rapid development of social networks, people become more and more willing to share moods via microblog platforms. In this paper, we propose an automatic stress detection method from cross-media microblog data. We construct a three-level framework to formulate the problem. We first obtain a set of low-level features from the tweets. Then we define and extract middle-level representations based on psychological and art theories: linguistic attributes from tweets' texts, visual attributes from tweets' images, and social attributes from tweets' comments, retweets and favorites. Finally, a Deep Sparse Neural Network is designed to learn the stress categories incorporating the cross-media attributes. Experiment results show that the proposed method is effective and efficient on detecting psychological stress from microblog data.

Sentiment analysis from large-scale networked data attracts increasing attention in recent years. Most previous works on sentiment prediction mainly focus on text or image data. However, voice is the most natural and direct way to express people's sentiments in real-time. With the rapid development of smart phone voice dialogue applications (e.g., Siri and Sogou Voice Assistant), the large-scale networked voice data can help us better quantitatively understand the sentimental world we live in. In this paper, we study the problem of sentiment prediction from large-scale networked voice data. In particular, we first investigate the data observations and underlying sentiment patterns in human-mobile voice communication. Then we propose a deep sparse neural network (DSNN) model to incorporate acoustic features, content information and geo-information to automatically predict sentiments. The effectiveness of the proposed model is verified by the experiments on a real dataset from Sogou Voice Assistant application.

It is of significant importance to detect and manage stress before it turns into severe problems. However, existing stress detection methods usually rely on psychological scales or physiological devices, making the detection complicated and costly. In this paper, we explore to automatically detect individuals' psychological stress via social media. Employing real online micro-blog data, we first investigate the correlations between users' stress and their tweeting content, social engagement and behavior patterns. Then we define two types of stress-related attributes: 1) low-level content attributes from a single tweet, including text, images and social interactions; 2) user-scope statistical attributes through their weekly micro-blog postings, leveraging information of tweeting time, tweeting types and linguistic styles. To combine content attributes with statistical attributes, we further design a convolutional neural network (CNN) with cross autoencoders to generate user-scope content attributes from low-level content attributes. Finally, we propose a deep neural network (DNN) model to incorporate the two types of user-scope attributes to detect users' psychological stress. We test the trained model on four different datasets from major micro-blog platforms including Sina Weibo, Tencent Weibo and Twitter. Experimental results show that the proposed model is effective and efficient on detecting psychological stress from micro-blog data. We believe our model would be useful in developing stress detection tools for mental health agencies and individuals.

The transmission control protocol (TCP) Friendly Rate Control (TFRC) protocol emulates a TCP-like slow start algorithm, which is known to cause two problems: (1) takes many round-trip times to reach the optimal operating point; (2) misleads the sender to send too many packets too quickly at the end of the slow start phase. In this paper, we propose a modified measurement-based slow start mechanism, called bandwidth estimate start (BE-start), to improve the start-up performance of the TFRC protocol. The BE-start employs an effective online bandwidth measurement technique to update the sending rate with appropriate value dynamically. By adapting to network conditions during the start-up phase, the sender is able to increase the sending rate quickly without incurring a risk of buffer overflow and multiple losses. Simulation experiments show that the BE-start can significantly improve the TFRC protocol performance under various bandwidths, buffer sizes and round-trip times. The method avoids both under-utilization due to premature slow start termination, as well as multiple losses due to increasing the sending rate too fast.

IT service continuity management (ITSCM) is very important to current organizations. It ensures that the organizations can reduce and limit tangible and intangible losses of business in the events of contingencies or disasters. However, current risk management approaches and methodologies cannot focus on ITSCM. Therefore, this paper presents a risk assessment method to evaluate the risk exposure of one service. Further, we show how to use the risk exposure metric to select the optimal proactive measures. The optimal measure set is achieved through a set of experimental results.