Fatigue is a significant safety concern across various domains, and accurate detection is vital. However, the commonly employed fine-grained labels (seconds-based) frequently inherit coarse-grained ...labels (minutes-based or more), inevitably introducing noise due to fatigue states’ dynamic and time-varying nature. Compared to noise in existing image tasks, fatigue noise is complex and diverse, with relatively ambiguous category boundaries. To address this issue, we propose a novel class-level fatigue noise-tolerant supervised comparative learning (cFNSCL) method that explores the data structure within each class, extracts trustworthy samples, and encourages the learning of distinguishable representations against noise. Specifically, supervised contrastive learning (SCL) is introduced to deal with complex and variable noise, and a dynamic noise-tolerant contrastive loss (DNCL) that incorporates a novel class-level confidence assessment mechanism (CCAM) is designed. CCAM selects high-confidence samples for DNCL learning, significantly alleviating SCL’s sensitivity to noise and enhancing the model’s tolerance to noise. Additionally, a novel class-level trustworthy sample extraction (cTSE) method from the perspective of inherited label categories to improve the model’s representative ability is proposed. Experimental results demonstrate the effectiveness of cFNSCL on both synthetic and real-world noisy datasets over some state-of-the-art methods. Specifically, it improves the accuracy by an average of 6.11% and 6.95% on the two real-world datasets, respectively.
Label Distribution Learning Geng, Xin
IEEE transactions on knowledge and data engineering,
2016-July-1, 2016-7-1, 20160701, Volume:
28, Issue:
7
Journal Article
Peer reviewed
Open access
Although multi-label learning can deal with many problems with label ambiguity, it does not fit some real applications well where the overall distribution of the importance of the labels matters. ...This paper proposes a novel learning paradigm named label distribution learning (LDL) for such kind of applications. The label distribution covers a certain number of labels, representing the degree to which each label describes the instance. LDL is a more general learning framework which includes both single-label and multi-label learning as its special cases. This paper proposes six working LDL algorithms in three ways: problem transformation, algorithm adaptation, and specialized algorithm design. In order to compare the performance of the LDL algorithms, six representative and diverse evaluation measures are selected via a clustering analysis, and the first batch of label distribution datasets are collected and made publicly available. Experimental results on one artificial and 15 real-world datasets show clear advantages of the specialized algorithms, which indicates the importance of special design for the characteristics of the LDL problem.
Label noises, categorized into closed-set noise and open-set noise, are prevalent in real-world scenarios and can seriously hinder the generalization ability of models. Identifying noise is ...challenging because noisy samples closely resemble true positives. Existing approaches often assume a single noise source, oversimplify closed-set noise, or treat open-set noise as toxic and eliminate it, resulting in limited practical effects. To address these issues, we present a novel approach named uncertainty-guided label correction with wavelet-transformed discriminative representation enhancement (Ultra), designed to mitigate the effects of mixed noise. Specifically, our approach considers a more practical noise setting. To achieve robust mixed-noise identification, we initially look into a learnable wavelet filter for obtaining discriminative features and filtering spurious cues automatically at the representation level. Subsequently, we introduce a two-fold uncertainty estimation to stably locate noise within the corrupted supervised signal level. These insights pave the way for a simple yet potent label correction technique, enabling comprehensive utilization of open-set noise, which can be rendered non-toxic in a specific manner, in contrast to harmful closed-set noise. Experimental validation on datasets with synthetic mixed noise, web noise corruption, and a real-world dataset confirms the effectiveness and generality of Ultra. Furthermore, our approach enhances the application of efficient techniques (e.g., supervised contrastive learning) within label noise scenarios.
Multi-label learning associates a given data instance with one or several class labels. A frequent problem with real life multi-label datasets is the lack of complete label information. Incomplete ...labels increase model complexity as the label correlation information is not reliable, resulting in a suboptimal multi-label classifier. Further, high dimensionality of multi-label datasets often introduces spurious feature-label dependencies. Thus, discovering label-specific features is imperative for efficient handling of high-dimensional data for multi-label learning with missing labels. To deal with the issues emerging from incomplete labels and high-dimensional input space, we propose a multi-label learning approach based on identifying the label-specific features and constraining them with a sparse global structure. The sparse structural constraint helps maintain the typical characteristics of the multi-label learning data. Instances are expressed as linear combination of label-specific features and the inter-relation guides the construction of model coefficients. The model also constructs supplementary label correlations to assist missing label recovery as part of the optimization problem. Empirical results on benchmark multi-label datasets highlight the effectiveness of the proposed method.
Einleitung: Die selektive Darstellung hirnversorgender Arterien ist eine wichtige differentialdiagnostische Methode der Neuroradiologie. Der Goldstandard ist die digitale Subtraktionsangiographie ...unter Einsatz von Kontrastmittel und Rontgenstrahlung. In dieser Studie wird eine nicht-invasive Methode der selektiven, kontrastmittelfreien Angiographie im MRT mittels Arterial Spin Labeling (ASL) prasentiert. Methoden: ASL ist eine Methode zur magnetischen Markierung einstromender Blutspins. Durch Subtraktion von Aufnahmen mit invertierten und nicht-invertierten Blutspins lassen sich Angiogramme erstellen. Eine modifizierte Methode (Superselektives ASL) erlaubt es zudem, nur das Blut einzelner Gefasse selektiv zu markieren. Zur dynamischen Darstellung des einstromenden Blutes wird die Wartezeit zwischen Inversion und Akquisition schrittweise vergrossert. Da dies die Gesamtmesszeit erhoht, wird zur Reduzierung der Akquisitionszeit die Keyhole-Technik eingesetzt, welche ublicherweise nur bei Kontrastmittelscans verwendet wird. In Simulationen wurden dazu Sequenzparameter optimiert (Keyhole Factor, Flipwinkel) und anschliessend in Phantom- und Probandenmessungen verifiziert. Scan-Parameter: Philips 3T Achieva, 400 msec Labeling, 3D T1-TFE Scan, FoV 200 x 200 mm.sup.2, Voxel: 0,9 x 0,9 x 0,9 mm.sup.3, 25degrees Flipwinkel, 100 ms Zeitauflosung, 5 min Messzeit. Ergebnisse: Der Blutfluss einzelner hirnversorgender Arterien (ACI, BA) konnte zeitaufgelost bis zu den Gefassabschnitten 4. Ordnung visualisiert werden. Schlussfolgerung: Superselektives ASL mit Keyhole-Akquisition erlaubt die Aufnahme gefassselektiver Angiogramme ohne Kontrastmittelgabe und in klinisch akzeptabler Messzeit.
Fragestellung: Pseudo-continuous Arterial Spin Labeling (pcASL) ist eine nicht-invasive, BOLD-basierte Methode zur Messung des zerebralen Blutflusses (CBF) mittels MRT. In dieser Studie wurde ...untersucht, inwieweit die regionale zerebrale Perfusion bei Patienten mit hochgradigen extrakraniellen Stenosen mittels pcASL charakterisiert werden kann. Methoden: Bei 50 Patienten (Alter 72 + or - 8,1; 16 weiblich) mit hochgradiger extrakranieller arterieller Stenose wurde der CBF mittels pcASL an einem 3 T MRT (label duration 1650 ms, post label delay 1525 ms) gemessen. Die so identifizierten Areale mit herabgesetztem CBF wurden mit denjenigen perfusionsgeminderten Arealen korreliert, die in der etablierten Perfusionsbildgebung (T2*-gewichtete DSC PWI) detektiert wurden (Pearson-Korrelationskoeffizient und Bland-Altman; n = 18). Die CBF-Werte wurden gemessen sowie die perfusionsgeminderten Volumina nach Versorgungsgebieten aufgeteilt. Der Schweregrad der Gefassstenosen wurde mit den mittels pcASL gemessenen CBF-Werten und Volumina korreliert (n = 17). Ergebnisse: Die Messung der relativen CBF in Arealen mit stenosiertem Versorgungsgefass in der globalen pcASL zeigte eine signifikante Korrelation mit den CBF-Werten gemessen mittels PWI (r = 0,72, P = 0,041). CBF-Lasionsvolumina gemessen mit pcASL korrelierten gut mit CBF-Lasionsvolumen gemessen mittels T2* PWI (r = 0,74; P < 0,01). Es bestand eine signifikante Korrelation zwischen verbliebenem Gefasslumen des stenosierten Gefasses und regionalem Perfusionsvolumen, gemessen mit pcASL (r = 0,82, P = 0,027). Schlussfolgerungen: pcASL ist eine valide und praktikable Methode zur nicht-invasiven Charakterisierung der regionalen zerebralen Perfusion bei Patienten mit hochgradiger arterieller Stenose.
Fragestellung: Die Abschatzung der Rezidiv-freien Uberlebenszeit (RFU) ist wichtig fur eine patientenmassgeschneiderte Therapie bei Gliompatienten. Ziel dieser Studie war, den mittels Arterial Spin ...Labeling (ASL) berechneten Tumorblutfluss als unabhangiger prognostischer Biomarker fur die RFU zu evaluieren. Methoden: Bei 24 Patienten mit primaren Gliomen wurden ASL-basierte Blutflussmessungen (Q2TIPS-PICORE pulsed ASL) des gesamten Tumors bei 3 T durchgefuhrt. Bei allen Patienten erfolgte eine histologische Diagnosesicherung. Receiver-operating-characteristic (ROC) Kurven zur Bestimmung des optimalen maximalen Tumorblutfluss (mTBF) Schwellenwertes fur die Prognose eines Rezidivs wurden angewandt. Die prognostische Wertigkeit von mTBF fur die RFU wurde mittels Kaplan-Meier und Cox proportional hazard Regressionsanalyse ermittelt. Ergebnisse: Aufgrund der oligodendroglialen Tumorkomponenten wurden 6 Patienten ausgeschlossen. Bei 18 Patienten wurden 2 WHO-Grad II, 2 WHO-Grad III und 14 WHOGrad IV Gliome diagnostiziert. Der optimale mTBF-Schwellenwert fur die Pradiktion eines Rezidivs lag bei 182 ml/ min/100 g Gewebe (Sensitivitat/Spezifitat: 83/100%>). Alle Tumore mit mTBF < 182 ml/min/100 g hatten signifikant langere RFU (772,5 + or - 290,9 Tage vs. 181,8 + or - 129,8 Tage bei Patienten mit mTBF greater than or equal to 82 ml/min/100 g) unabhangig von WHO-Grad (P = 0,0012). Bei der multivariaten Cox-Regressionsanalyse erwiesen sich nur das Resektionsausmass (P=0,04) und mTBF (P = 0,0046) als unabhangige prognostische Faktoren fur die RFU. Schlussfolgerungen: ASL-basierte mTBF-Werte bieten einen neuen, nicht-invasiven Biomarker fur die RFU-Prognose in Gliomen, unabhangig vom histologischen Grading.
Brazil is currently debating the implementation of front-of-package labels. This study tested if Warning labels (WLs) improved consumer understanding, perceptions, and purchase intentions compared to ...Traffic-Light labels (TLLs) in 1607 Brazilian adults.
In this online, randomized controlled experiment participants saw images of 10 products and answered questions twice-once in a no-label, control condition and then again in a randomly assigned label condition. The relative differences in responses between WLs and TLLs between control and label conditions were estimated using one-way ANOVAs or Chi-square tests.
Presenting WLs on products compared to TLLs helped participants: (i) improve their understanding of excess nutrient content (27.0% versus 8.2%,
< 0.001); (ii) improve their ability to identify the healthier product (24.6% versus 3.3%,
< 0.001); (iii) decrease perceptions of product healthfulness; and (iv) correctly identify healthier products (14.0% versus 6.9%,
< 0.001), relative to the control condition. With WLs, there was also an increase in the percentage of people: (v) expressing an intention to purchase the relatively healthier option (16.1% versus 9.8%,
< 0.001); and (vi) choosing not to buy either product (13.0% versus 2.9%,
< 0.001), relative to the control condition. The participants in the WL condition had significantly more favorable opinions of the labels compared to those in the TLL group.
WLs would be more effective at improving consumer food choices.
Health warning labels (HWLs) using images and text to depict the negative health consequences of tobacco consumption are effective and acceptable for changing smoking-related outcomes. There is ...currently limited evidence concerning their potential use for reducing consumption of alcoholic drinks and energy-dense foods. The aim of this research was to describe the potential effectiveness and acceptability of image-and-text (also known as pictorial or graphic) HWLs applied to: i. alcoholic drinks and ii. energy-dense snack foods.
Two online studies were conducted using between-subjects designs with general population samples. Participants rated one of 21 image-and-text HWLs on alcoholic drinks (n = 5528), or one of 18 image-and-text HWLs on energy-dense snacks (n = 4618). HWLs comprised a graphic image with explanatory text, depicting, respectively, seven diseases linked to excess alcohol consumption, and six diseases linked to excess energy intake. Diseases included heart disease and various cancers. Outcomes were negative emotional arousal, desire to consume the labelled product, and acceptability of the label. Free-text comments relating to HWLs were content analysed.
For both alcoholic drinks and energy-dense snacks, HWLs depicting bowel cancer generated the highest levels of negative emotional arousal and lowest desire to consume the product, but were the least acceptable. Acceptability was generally low for HWLs applied to alcohol, with 3 of 21 rated as acceptable, and was generally high for snacks, with 13 of 18 rated as acceptable. The majority of free-text comments expressed negative reactions to HWLs on alcohol or energy-dense snacks.
Image-and-text health warning labels depicting bowel cancer showed greatest potential for reducing selection and consumption of alcoholic drinks and energy-dense snacks, although they were the least acceptable. Laboratory and field studies are needed to assess their impact on selection and consumption.
Policy makers around the world are facing serious challenges in controlling citizens' obesity and healthiness, hence, they devote increased attention to the development of tools that communicate ...easily processable nutrition information. Front‐of‐package nutrition labels are one of such tools and have been used to signal the extent to which food items contain potentially unhealthy ingredients such as sugar or fat. In this research, we focus on sugar cues on three different food categories to investigate their impact on consumer choice. We compare two labels, one already used (traffic‐light) and one never used (sugar teaspoon): sugar teaspoons prove to be more effective than the previously used traffic‐lights in healthy product choices, but only for specific food categories. In two experiments, we find that sugar teaspoon labels indicating sugar content, as opposed to traffic‐light labels, are more effective in signaling sugar levels and thus, in helping consumers making healthier choices. We find that this is particularly relevant for food categories that have a simpler ingredient composition (i.e., whose healthiness relies more heavily on sugar). We finally propose processing fluency as the mechanism for the relation between sugar signals and product choices.