Unsupervised learning represents one of the most interesting challenges in computer vision today. The task has an immense practical value with many applications in artificial intelligence and ...emerging technologies, as large quantities of unlabeled images and videos can be collected at low cost. In this paper, we address the unsupervised learning problem in the context of segmenting the main foreground objects in single images. We propose an unsupervised learning system, which has two pathways, the teacher and the student, respectively. The system is designed to learn over several generations of teachers and students. At every generation the teacher performs unsupervised object discovery in videos or collections of images and an automatic selection module picks up good frame segmentations and passes them to the student pathway for training. At every generation multiple students are trained, with different deep network architectures to ensure a better diversity. The students at one iteration help in training a better selection module, forming together a more powerful teacher pathway at the next iteration. In experiments, we show that the improvement in the selection power, the training of multiple students and the increase in unlabeled data significantly improve segmentation accuracy from one generation to the next. Our method achieves top results on three current datasets for object discovery in video, unsupervised image segmentation and saliency detection. At test time, the proposed system is fast, being one to two orders of magnitude faster than published unsupervised methods. We also test the strength of our unsupervised features within a well known transfer learning setup and achieve competitive performance, proving that our unsupervised approach can be reliably used in a variety of computer vision tasks.
Heterozygous gain-of-kinase function variants in
LRRK2
(leucine-rich repeat kinase 2) cause 1–2% of all cases of Parkinson’s disease (PD) albeit with incomplete and age-dependent penetrance. All ...pathogenic LRRK2 mutations reside within the two catalytic domains of LRRK2—either in its kinase domain (e.g. G2019S) with modest effect or its ROC-COR GTPase domain (e.g. R1441G/H) with large effect on LRRK2 kinase activity. We have previously reported assays to interrogate LRRK2 kinase pathway activity in human bio-samples measuring phosphorylation of its endogenous substrate Rab10, that mirrors LRRK2 kinase activation status. Here, we isolated neutrophils from fresh peripheral blood from 101 participants including 42 LRRK2 mutation carriers (21 with the G2019S and 21 with the R1441G mutations), 27 patients with idiopathic PD, and 32 controls. Using a dual approach, LRRK2 dependent Rab10 phosphorylation at Threonine 73 (pRab10
Thr73
) was measured by quantitative multiplexed immunoblotting for pRab10
Thr73
/total Rab10 as well as targeted mass-spectrometry for absolute pRab10
Thr73
occupancy. We found a significant over fourfold increase in pRab10
Thr73
phosphorylation in carriers of the LRRK2 R1441G mutation irrespective of clinical disease status. The effect of the LRRK2 G2019S mutation did not reach statistical significance. Furthermore, we show that LRRK2 phosphorylation at Serine 935 is not a marker for LRRK2 kinase activity in human neutrophils. When analysing pRab10
Thr73
phosphorylation in post-mortem brain samples, we observed overall high variability irrespective of clinical and LRRK2 mutation status and attributed this mainly to the adverse effect of the peri- and post-mortem period on the stability of posttranslational modifications such as protein phosphorylation. Overall, in vivo LRRK2 dependent pRab10
Thr73
phosphorylation in human peripheral blood neutrophils is a specific, robust and promising biomarker for significant LRRK2 kinase hyperactivation, as with the LRRK2 R1441G mutation. Additional readouts and/or assays may be needed to increase sensitivity to detect modest LRRK2 kinase activation, as with the LRRK2 G2019S mutation. Our assays could be useful for patient stratification and target engagement studies for LRRK2 kinase inhibitors.
Parkinson's disease (PD) is characterized by a great clinical heterogeneity. Nevertheless, the biological drivers of this heterogeneity have not been completely elucidated and are likely to be ...complex, arising from interactions between genetic, epigenetic, and environmental factors. Despite this heterogeneity, the clinical patterns of monogenic forms of PD have usually maintained a good clinical correlation with each mutation once a sufficient number of patients have been studied. Mutations in LRRK2 are the most commonly known genetic cause of autosomal dominant PD known to date. Furthermore, recent genome-wide association studies have revealed variations in LRRK2 as significant risk factors also for the development of sporadic PD. The LRRK2-R1441G mutation is especially frequent in the population of Basque ascent based on a possible founder effect, being responsible for almost 50% of cases of familial PD in our region, with a high penetrance. Curiously, Lewy bodies, considered the neuropathological hallmark of PD, are absent in a significant subset of LRRK2-PD cases. Indeed, these cases appear to be associated with a less aggressive primarily pure motor phenotype. The aim of our research is to examine the clinical phenotype of R1441G-PD patients, more homogeneous when we compare it with sporadic PD patients or with patients carrying other LRRK2 mutations, and reflect on the value of the observed correlation in the genetic forms of PD. The clinical heterogeneity of PD leads us to think that there may be as many different diseases as the number of people affected. Undoubtedly, genetics constitutes a relevant key player, as it may significantly influence the phenotype, with differences according to the mutation within the same gene, and not only in familial PD but also in sporadic forms. Thus, extending our knowledge regarding genetic forms of PD implies an expansion of knowledge regarding sporadic forms, and this may be relevant due to the future therapeutic implications of all forms of PD.
Elevated urine bis(monoacylglycerol)phosphate (BMP) levels have been found in gain-of-kinase function LRRK2 G2019S mutation carriers. Here, we have expanded urine BMP analysis to other Parkinson's ...disease (PD) associated mutations and found them to be consistently elevated in carriers of LRRK2 G2019S and R1441G/C as well as VPS35 D620N mutations. Urine BMP levels are promising biomarkers for patient stratification and potentially target engagement in clinical trials of emerging targeted PD therapies.
Parkinson´s disease (PD) is a common neurodegenerative movement disorder and leucine-rich repeat kinase 2 (LRRK2) is a promising therapeutic target for disease intervention. However, the ability to ...stratify patients who will benefit from such treatment modalities based on shared etiology is critical for the success of disease-modifying therapies. Ciliary and centrosomal alterations are commonly associated with pathogenic LRRK2 kinase activity and can be detected in many cell types. We previously found centrosomal deficits in immortalized lymphocytes from G2019S-LRRK2 PD patients. Here, to investigate whether such deficits may serve as a potential blood biomarker for PD which is susceptible to LRKK2 inhibitor treatment, we characterized patient-derived cells from distinct PD cohorts. We report centrosomal alterations in peripheral cells from a subset of early-stage idiopathic PD patients which is mitigated by LRRK2 kinase inhibition, supporting a role for aberrant LRRK2 activity in idiopathic PD. Centrosomal defects are detected in R1441G-LRRK2 and G2019S-LRRK2 PD patients and in non-manifesting LRRK2 mutation carriers, indicating that they accumulate prior to a clinical PD diagnosis. They are present in immortalized cells as well as in primary lymphocytes from peripheral blood. These findings indicate that analysis of centrosomal defects as a blood-based patient stratification biomarker may help nominate idiopathic PD patients who will benefit from LRRK2-related therapeutics.
In recent years, considerable progress on the task of text-video retrieval has been achieved by leveraging large-scale pretraining on visual and audio datasets to construct powerful video encoders. ...By contrast, despite the natural symmetry, the design of effective algorithms for exploiting large-scale language pretraining remains under-explored. In this work, we are the first to investigate the design of such algorithms and propose a novel generalized distillation method, TeachText, which leverages complementary cues from multiple text encoders to provide an enhanced supervisory signal to the retrieval model. Moreover, we extend our method to video side modalities and show that we can effectively reduce the number of used modalities at test time without compromising performance. Our approach advances the state of the art on several video retrieval benchmarks by a significant margin and adds no computational overhead at test time. Last but not least, we show an effective application of our method for eliminating noise from retrieval datasets. Code and data can be found at https://www.robots.ox.ac.uk/˜vgg/research/teachtext/.
Unsupervised learning from visual data is one of the most difficult challenges in computer vision. It is essential for understanding how visual recognition works. Learning from unsupervised input has ...an immense practical value, as huge quantities of unlabeled videos can be collected at low cost. Here we address the task of unsupervised learning to detect and segment foreground objects in single images. We achieve our goal by training a student pathway, consisting of a deep neural network that learns to predict, from a single input image, the output of a teacher pathway that performs unsupervised object discovery in video. Our approach is different from the published methods that perform unsupervised discovery in videos or in collections of images at test time. We move the unsupervised discovery phase during the training stage, while at test time we apply the standard feed-forward processing along the student pathway. This has a dual benefit: firstly, it allows, in principle, unlimited generalization possibilities during training, while remaining fast at testing. Secondly, the student not only becomes able to detect in single images significantly better than its unsupervised video discovery teacher, but it also achieves state of the art results on two current benchmarks, YouTube Objects and Object Discovery datasets. At test time, our system is two orders of magnitude faster than other previous methods.
Cross Modal Retrieval with Querybank Normalisation Bogolin, Simion-Vlad; Croitoru, Ioana; Jin, Hailin ...
2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),
2022-June
Conference Proceeding
Profiting from large-scale training datasets, advances in neural architecture design and efficient inference, joint embeddings have become the dominant approach for tackling cross-modal retrieval. In ...this work we first show that, despite their effectiveness, state-of-the-art joint embeddings suffer significantly from the longstanding "hubness problem" in which a small number of gallery embeddings form the nearest neighbours of many queries. Drawing inspiration from the NLP literature, we formulate a simple but effective framework called Querybank Normalisation (QB-NORM) that re-normalises query similarities to account for hubs in the embedding space. QB-NORM improves retrieval performance without requiring retraining. Differently from prior work, we show that QB-NORM works effectively without concurrent access to any test set queries. Within the QB-NORM framework, we also propose a novel similarity normalisation method, the Dynamic Inverted Softmax, that is significantly more robust than existing approaches. We showcase QB-NORM across a range of cross modal retrieval models and benchmarks where it consistently enhances strong baselines beyond the state of the art. Code is available at https://vladbogo.github.io/QB-Norm/.
Background and purpose
Dominantly inherited GAA repeat expansions in the fibroblast growth factor 14 (FGF14) gene have recently been shown to cause spinocerebellar ataxia 27B (SCA27B). We aimed to ...study the frequency and phenotype of SCA27B in a cohort of patients with unsolved late‐onset cerebellar ataxia (LOCA). We also assessed the frequency of SCA27B relative to other genetically defined LOCAs.
Methods
We recruited a consecutive series of 107 patients with LOCA, of whom 64 remained genetically undiagnosed. We screened these 64 patients for the FGF14 GAA repeat expansion. We next analysed the frequency of SCA27B relative to other genetically defined forms of LOCA in the cohort of 107 patients.
Results
Eighteen of 64 patients (28%) carried an FGF14 (GAA)≥250 expansion. The median (range) age at onset was 62.5 (39–72) years. The most common clinical features included gait ataxia (100%) and mild cerebellar dysarthria (67%). In addition, episodic symptoms and downbeat nystagmus were present in 39% (7/18) and 37% (6/16) of patients, respectively. SCA27B was the most common cause of LOCA in our cohort (17%, 18/107). Among patients with genetically defined LOCA, SCA27B was the main cause of pure ataxia, RFC1‐related disease of ataxia with neuropathy, and SPG7 of ataxia with spasticity.
Conclusion
We showed that SCA27B is the most common cause of LOCA in our cohort. Our results support the use of FGF14 GAA repeat expansion screening as a first‐tier genetic test in patients with LOCA.