Deep learning with noisy labels is practically challenging, as the capacity of deep models so high that they can totally memorize these sooner or later during training. Nonetheless, recent studies on memorization effects neural networks show would first training data clean and then those labels. Therefore in this paper, we propose a new paradigm called ''Co-teaching'' for combating Namely, train two simultaneously, let them teach each other given every mini-batch: firstly, network feeds...

In daily life, one of the most common social tasks we perform is to recognize faces. However, relation between face recognition ability and activities largely unknown. Here ask whether individuals with better skills are also at recognizing We found that extraverts who have correctly recognized more faces than introverts. this advantage was absent when were asked non-social stimuli (e.g., flowers). particular, underlying facet makes a recognizer mainly due gregariousness measures degree...

Partial-label learning (PLL) is a typical weakly supervised problem, where each training instance equipped with set of candidate labels among which only one the true label. Most existing methods elaborately designed objectives as constrained optimizations that must be solved in specific manners, making their computational complexity bottleneck for scaling up to big data. The goal this paper propose novel framework PLL flexibility on model and optimization algorithm. More specifically, we...

Partial-label learning (PLL) is a multi-class classification problem, where each training example associated with set of candidate labels. Even though many practical PLL methods have been proposed in the last two decades, there lacks theoretical understanding consistency those methods-none hitherto possesses generation process label sets, and then it still unclear why such method works on specific dataset when may fail given different dataset. In this paper, we propose first model develop...

Positive-unlabeled (PU) learning deals with the binary classification problem when only positive (P) and unlabeled (U) data are available, without negative (N) data. Existing PU methods perform well on balanced dataset. However, in real applications such as financial fraud detection or medical diagnosis, always imbalanced. It remains unclear whether existing can imbalanced In this paper, we explore propose a general objective for targeting specially at By objective, state-of-the-art based...

Abstract Medical data refers to health-related information associated with regular patient care or as part of a clinical trial program. There are many categories such data, imaging bio-signal electronic health records (EHR), and multi-modality medical data. With the development deep neural networks in last decade, emerging pre-training paradigm has become dominant that it significantly improved machine learning methods′ performance data-limited scenario. In recent years, studies domain have...

Abstract Multivariate time series (MTS) data are vital for various applications, particularly in machine learning tasks. However, challenges such as sensor failures can result irregular and misaligned with missing values, thereby complicating their analysis. While recent advancements use graph neural networks (GNNs) to manage these Irregular Time Series (IMTS) data, they generally require a reliable structure, either pre-existing or inferred from adequate properly capture node correlations....

Label distribution learning (LDL) assumes labels can be associated to an instance some degree, thus it learn the relevance of a label particular instance. Although LDL has got successful practical applications, one problem with existing methods is that they are designed for data \emph{complete} supervised information, while in reality, annotation information may \emph{incomplete}, because assigning each real value indicate its association will result large cost labor and time. In this paper,...

Medication recommendation is a significant healthcare application due to its promise in effectively prescribing medications. Avoiding fatal side effects related Drug-Drug Interaction (DDI) among the critical challenges. Most existing methods try mitigate problem by providing models with extra DDI knowledge, making complicated. While treating all patients different properties as single cohort would put forward strict requirements on models' generalization performance. In pursuit of valuable...

Partial-label learning (PLL) utilizes instances with PLs, where a PL includes several candidate labels but only one is the true label (TL). In PLL, identification-based strategy (IBS) purifies each on fly to select (most likely) TL for training; average-based (ABS) treats all equally training and let trained models be able predict TL. Although PLL research has focused IBS better performance, ABS also worthy of study since modern behaves like in beginning prepare purification selection. this...

Feature missing is a serious problem in many applications, which may lead to low quality of training data and further significantly degrade the learning performance. While feature acquisition usually involves special devices or complex process, it expensive acquire all values for whole dataset. On other hand, features be correlated with each other, some recovered from others. It thus important decide are most informative recovering as well improving In this paper, we try train an effective...

Deep learning with noisy labels is practically challenging, as the capacity of deep models so high that they can totally memorize these sooner or later during training. Nonetheless, recent studies on memorization effects neural networks show would first training data clean and then those labels. Therefore in this paper, we propose a new paradigm called Co-teaching for combating Namely, train two simultaneously, let them teach each other given every mini-batch: firstly, network feeds forward...

Multi-label learning methods assign multiple labels to one object. In practice, in addition differentiating relevant from irrelevant ones, it is often desired rank the for an object, whereas rankings of are not important. Such a requirement, however, cannot be met because most existing were designed optimize criteria, yet there no criterion which encodes aforementioned requirement. this paper, we present new criterion, Pro Loss, concerning prediction on all as well only labels. We then...

Moth-Flame Optimization (MFO) algorithm is a widely used nature-inspired optimization algorithm. However, for some complex problems, such as high dimensional and multimodal the MFO may fall into local optimal solution. Hence, in this paper an ameliorated (AMFO) presented to improve solution quality global capability. The key features of proposed are Gaussian mutation produce flames modified position updating mechanism moths, which can ability jump out optimum solutions. In addition,...

In e-commerce advertising, the ad platform usually relies on auction mechanisms to optimize different performance metrics, such as user experience, advertiser utility, and revenue. However, most of state-of-the-art only focus optimizing a single metric, e.g., either social welfare or revenue, are not suitable for advertising with various, dynamic, difficult estimate, even conflicting metrics. this paper, we propose new mechanism called Deep GSP auction, which leverages deep learning design...

CUR matrix decomposition computes the low rank approximation of a given by using actual rows and columns matrix. It has been very useful tool for handling large matrices. One limitation with existing algorithms is that they need an access to {\it full} matrix, requirement can be difficult fulfill in many real world applications. In this work, we alleviate developing algorithm partially observed particular, proposed target based on (i) randomly sampled columns, (ii) subset entries are from...

Multi-label learning methods assign multiple labels to one object. In practice, in addition differentiating relevant from irrelevant ones, it is often desired rank for an object, whereas the ranking of not important. Thus, we require algorithm do classification and simultaneously. Such a requirement, however, cannot be met because most existing were designed optimize criteria, yet there no criterion which encodes aforementioned requirement. this paper, present new criterion, PRO LOSS,...

This paper proposes a fixed-time backstepping control strategy based on adaptive super-twisting disturbance observers (ASTDOs) for class of non-integral cascade high-order uncertain nonlinear systems. First, the ASTDOs are designed to estimate system disturbances in fixed time, and method relaxes assumption about disturbances, which is that upper bound first second derivatives unknown. Then we present via an improved distributed fast terminal sliding mode, not only guarantees mode surface...