High-performance molecular property prediction is one of the essential problems in many research fields. Although Deep Learning models (DLMs) have the strong potential to predict molecular property, the search space o...
High-performance molecular property prediction is one of the essential problems in many research fields. Although Deep Learning models (DLMs) have the strong potential to predict molecular property, the search space of their neural structure and hyperparameters is too huge to enumerate all possibilities, which implies that one will be difficult to get a optimal configuration solution, or even build a DLM with poor performance. To tackle the aforementioned problem, we first abstractly map various types of supervised DLMs for chemical molecular property prediction under a unified molecular DLM conceptualization. Then, we develop a self-learning molecular DLMs construction and optimization framework, called GEP-DL4Mol, based on the unified supervised molecular DLM conceptualization and Gene Expression Programming. Experiments on eight benchmark datasets including 19 property prediction tasks were conducted to evaluate the performance of the proposed method. Experimental results show that the proposed method outperforms other methods.
The Area Under the ROC Curve (AUC) is a widely employed metric in long-tailed classification scenarios. Nevertheless, most existing methods primarily assume that training and testing examples are drawn i.i.d. from the...
The Area Under the ROC Curve (AUC) is a widely employed metric in long-tailed classification scenarios. Nevertheless, most existing methods primarily assume that training and testing examples are drawn i.i.d. from the same distribution, which is often unachievable in practice. Distributionally Robust Optimization (DRO) enhances model performance by optimizing it for the local worst-case scenario, but directly integrating AUC optimization with DRO results in an intractable optimization problem. To tackle this challenge, methodically we propose an instance-wise surrogate loss of Distributionally Robust AUC (DRAUC) and build our optimization framework on top of it. Moreover, we highlight that conventional DRAUC may induce label bias, hence introducing distribution-aware DRAUC as a more suitable metric for robust AUC learning. Theoretically, we affirm that the generalization gap between the training loss and testing error diminishes if the training set is sufficiently large. Empirically, experiments on corrupted benchmark datasets demonstrate the effectiveness of our proposed method. Code is available at: https://***/EldercatSAM/DRAUC.
The dilemma between plasticity and stability arises as a common challenge for incremental learning. In contrast, the human memory system is able to remedy this dilemma owing to its multilevel memory structure, which m...
The dilemma between plasticity and stability arises as a common challenge for incremental learning. In contrast, the human memory system is able to remedy this dilemma owing to its multilevel memory structure, which motivates us to propose a Bilevel Memory system with Knowledge Projection (BMKP) for incremental learning. BMKP decouples the functions of learning and remembering via a bilevel-memory design: a working memory responsible for adaptively model learning, to ensure plasticity; a long-term memory in charge of enduringly storing the knowledge incorporated within the learned model, to guarantee stability. However, an emerging issue is how to extract the learned knowledge from the working memory and assimilate it into the long-term memory. To approach this issue, we reveal that the parameters learned by the working memory are actually residing in a redundant high-dimensional space, and the knowledge incorporated in the model can have a quite compact representation under a group of pattern basis shared by all incremental learning tasks. Therefore, we propose a knowledge projection process to adaptively maintain the shared basis, with which the loosely organized model knowledge of working memory is projected into the compact representation to be remembered in the long-term memory. We evaluate BMKP on CIFAR-10, CIFAR-100, and Tiny-ImageNet. The experimental results show that BMKP achieves state-of-the-art performance with lower memory usage 1 1 The code is available at https://***/SunWenJu123/BMKP.
Real-world datasets are typically imbalanced in the sense that only a few classes have numerous samples, while many classes are associated with only a few samples. As a result, a naïve ERM learning process will b...
Real-world datasets are typically imbalanced in the sense that only a few classes have numerous samples, while many classes are associated with only a few samples. As a result, a naïve ERM learning process will be biased towards the majority classes, making it difficult to generalize to the minority classes. To address this issue, one simple but effective approach is to modify the loss function to emphasize the learning on minority classes, such as re-weighting the losses or adjusting the logits via class-dependent terms. However, existing generalization analysis of such losses is still coarse-grained and fragmented, failing to explain some empirical results. To bridge this gap, we propose a novel technique named data-dependent contraction to capture how these modified losses handle different classes. On top of this technique, a fine-grained generalization bound is established for imbalanced learning, which helps reveal the mystery of re-weighting and logit-adjustment in a unified manner. Furthermore, a principled learning algorithm is developed based on the theoretical insights. Finally, the empirical results on benchmark datasets not only validate the theoretical results but also demonstrate the effectiveness of the proposed method.
Combination of multi-modal PET-CT imaging for lung tumor segmentation is significant for clinical treatment. Existing methods have not fully considered the impact of noise in PET-CT on the multi-modal interaction. To ...
详细信息
ISBN:
(数字)9798350390155
ISBN:
(纸本)9798350390162
Combination of multi-modal PET-CT imaging for lung tumor segmentation is significant for clinical treatment. Existing methods have not fully considered the impact of noise in PET-CT on the multi-modal interaction. To address this, we propose a novel Attention in Attention Network (AiANet). AiANet can mutually learn multi-modal characteristics for segmentation through its cross-learning modules. Within the cross-learning module, we introduce two nested-attention blocks, namely Attention in Self-Attention (AiSA) and Attention in Cross-Attention (AiCA), for multi-scale feature enhancement and multi-modal feature interaction. Importantly, since traditional attention weights calculated solely or unilaterally based on PET or CT can be vulnerable to the inevitable noisy information, we embed a novel Attention in Attention (AiA) module into AiSA and AiCA. The AiA module can seek cross-modal consensus for attention weights to alleviate their noise. Experimental results on clinical PET-CT data of lung cancer demonstrate the superiority of our method.
Visual-language models based on CLIP have shown remarkable abilities in general few-shot image classification. However, their performance drops in specialized fields such as healthcare or agriculture, because CLIP'...
详细信息
High-throughput proteomics based on mass spectrometry(MS) analysis has permeated biomedical science and propelled numerous research projects. p Find 3 is a database search engine for high-speed and in-depth proteomi...
详细信息
High-throughput proteomics based on mass spectrometry(MS) analysis has permeated biomedical science and propelled numerous research projects. p Find 3 is a database search engine for high-speed and in-depth proteomics data analysis. p Find 3 features a swift open search workflow that is adept at uncovering less obvious information such as unexpected modifications or mutations that would have gone unnoticed using a conventional data analysis pipeline. In this protocol, we provide step-by-step instructions to help users mastering various types of data analysis using p Find 3 in conjunction with p Parse for data pre-processing and if needed, p Quant for quantitation. This streamlined p Parse-p Findp Quant workflow offers exceptional sensitivity, precision, and speed. It can be easily implemented in any laboratory in need of identifying peptides, proteins, or post-translational modifications, or of quantitation based on15N-labeling, SILAC-labeling, or TMT/i TRAQ labeling.
Cooperative behavior has been of great concern in evolutionary game researches because of its important role in social life and natural evolution. Since memory can greatly influence the emergence of cooperative behavi...
Cooperative behavior has been of great concern in evolutionary game researches because of its important role in social life and natural evolution. Since memory can greatly influence the emergence of cooperative behavior, memory-based research on evolutionary games has proliferated in recent years. In the present paper, inspired by the well-known “Stanford marshmallow experiment” (SME), we proposed a multi-memory mechanism in Snowdrift game (SDG) on spatial lattice networks whose core lies in what follows: Players are divided into two groups according to individual self-control; different memory lengths are set by group and by time in a single realization. The simulation results show that our proposed mechanism promotes cooperation level of the considered evolutionary game. More specifically, as memory length increases to an intermediate length, the cooperation level increases as well. For mechanism with short memory length, the effect of population structure is obvious and small values of adjusting factor can intensively facilitate high-frequency cooperative behavior. These findings may be helpful to understand the connection between population heterogeneity and the emergence of high cooperation.
Conductive Ti_(3)C_(2)T_(x)MXenes have been widely investigated for the construction of flexible and highly-sensitive pressure *** the inevitable oxidation of solution-processed MXene has been recognized,the effect of...
详细信息
Conductive Ti_(3)C_(2)T_(x)MXenes have been widely investigated for the construction of flexible and highly-sensitive pressure *** the inevitable oxidation of solution-processed MXene has been recognized,the effect of the irreversible oxidation of MXene on its electrical conductivity and sensing properties is yet to be ***,we construct a highly-sensitive and degradable piezoresistive pressure sensor by coating Ti_(3)C_(2)T_(x)MXene flakes with different degrees of in situ oxidation onto paper substrates using the dipping-drying *** situ oxidation can tune the intrinsic resistance and expand the interlayer distance of MXene *** partially oxidized MXene-based piezoresistive pressure sensor exhibits a high sensitivity of 28.43 kPa^(-1),which is greater than those of pristine MXene,over-oxidized MXene,and state-of-the-art paper-based pressure ***,these sensors exhibit a short response time of 98.3 ms,good durability over 5000 measurement cycles,and a low force detection limit of 0.8 ***,MXene-based sensing elements are easily degraded and environmentally *** MXene-based pressure sensor shows promise for practical applications in tracking body movements,sports coaching,remote health monitoring,and human–computer interactions.
Multi-label Pedestrian Attribute Recognition (PAR) involves identifying a series of semantic attributes in person images. Existing PAR solutions typically rely on CNN as the backbone network to extract pedestrian feat...
详细信息
Multi-label Pedestrian Attribute Recognition (PAR) involves identifying a series of semantic attributes in person images. Existing PAR solutions typically rely on CNN as the backbone network to extract pedestrian features. Unfortunately, CNNs process only one adjacent region at a time, resulting in the disappearance of long-range relations between different attribute-specific regions. To address this limitation, we adopt the Vision Transformer (ViT) instead of CNN as the backbone for PAR, aiming to build long-range relations and extract more robust features. However, PAR suffers from an inherent attribute imbalance issue, causing ViT to naturally focus more on attributes that appear frequently in the training set and ignore some pedestrian attributes that appear less. The native features extracted by ViT are not able to tolerate the imbalance attribute distribution issue. To tackle this issue, we propose a novel component and a dual-level loss: the Selective Feature Activation Method (SFAM), the Orthogonal Feature Activation Loss (OFALoss), and Orthogonal Weight Regularization Loss (OWRLoss). SFAM smartly suppresses the more informative attribute-specific features, thus compelling the PAR model to pay greater attention to attribute-specific regions that are often overlooked. The proposed OFALoss enforces an orthogonal constraint on the original feature extracted by ViT and the suppressed features from SFAM, promoting the comprehensiveness of feature representation in each attribute-specific region. Furthermore, OWRLoss is employed for decreasing correlations among entries of the last shared classification layer, which can alleviate the highly correlated of weight vectors caused by non-uniform distribution. This can prevent excessive mutual interference among different attributes during attribute recognition. Our model-agnostic approach is plug-and-play, requiring no additional training parameters in the training process. We conduct experiments on several benchmark P
暂无评论