Gotowa bibliografia na temat „Discriminative Encoder (DisCoder)”

Myxobacteria are an example of how single-cell individuals can transition into multicellular life by an aggregation strategy. For these and all organisms that consist of social groups of cells, discrimination against, and exclusion of, nonself is critical. In myxobacteria, TraA is a polymorphic cell surface receptor that identifies kin by homotypic binding, and in so doing exchanges outer membrane (OM) proteins and lipids between cells with compatible receptors. However, TraA variability alone is not sufficient to discriminate against all cells, as traA allele diversity is not necessarily high among local strains. To increase discrimination ability, myxobacteria include polymorphic OM lipoprotein toxins called SitA in their delivered cargo, which poison recipient cells that lack the cognate, allele-specific SitI immunity protein. We previously characterized 3 SitAI toxin/immunity pairs that belong to 2 families. Here, we discover 4 additional SitA families. Each family is unique in sequence, but share the characteristic features of SitA: OM-associated toxins delivered by TraA. We demonstrate that, within a SitA family, C-terminal nuclease domains are polymorphic and often modular. Remarkably, sitA loci are strikingly numerous and diverse, with most genomes possessing >30 and up to 83 distinct sitAI loci. Interestingly, all SitA protein families are serially transferred between cells, allowing a SitA inhibitor cell to poison multiple targets, including cells that never made direct contact. The expansive suites of sitAI loci thus serve as identify barcodes to exquisitely discriminate against nonself to ensure populations are genetically homogenous to conduct cooperative behaviors.

The key of Weakly Supervised Fine-grained Image Classification (WFGIC) is how to pick out the discriminative regions and learn the discriminative features from them. However, most recent WFGIC methods pick out the discriminative regions independently and utilize their features directly, while neglecting the facts that regions' features are mutually semantic correlated and region groups can be more discriminative. To address these issues, we propose an end-to-end Graph-propagation based Correlation Learning (GCL) model to fully mine and exploit the discriminative potentials of region correlations for WFGIC. Specifically, in discriminative region localization phase, a Criss-cross Graph Propagation (CGP) sub-network is proposed to learn region correlations, which establishes correlation between regions and then enhances each region by weighted aggregating other regions in a criss-cross way. By this means each region's representation encodes the global image-level context and local spatial context simultaneously, thus the network is guided to implicitly discover the more powerful discriminative region groups for WFGIC. In discriminative feature representation phase, the Correlation Feature Strengthening (CFS) sub-network is proposed to explore the internal semantic correlation among discriminative patches' feature vectors, to improve their discriminative power by iteratively enhancing informative elements while suppressing the useless ones. Extensive experiments demonstrate the effectiveness of proposed CGP and CFS sub-networks, and show that the GCL model achieves better performance both in accuracy and efficiency.

Inspired by the many reported successful applications of DNA-encoded chemical libraries in drug discovery projects with protein targets, we decided to apply this platform to nucleic acid targets. We used a 120-billion-compound set of 33 distinct DNA-encoded chemical libraries and affinity-mediated selection to discover binders to a panel of DNA targets. Here, we report the successful discovery of small molecules that specifically interacted with DNA G-quartets, which are stable structural motifs found in G-rich regions of genomic DNA, including in the promoter regions of oncogenes. For this study, we chose the G-quartet sequence found in the c-myc promoter as a primary target. Compounds enriched using affinity-mediated selection against this target demonstrated high-affinity binding and high specificity over DNA sequences not containing G-quartet motifs. These compounds demonstrated a moderate ability to discriminate between different G-quartet motifs and also demonstrated activity in a cell-based assay, suggesting direct target engagement in the cell. DNA-encoded chemical libraries and affinity-mediated selection are uniquely suited to discover binders to targets that have no inherent activity outside of a cellular context, and they may also be of utility in other nucleic acid structural motifs.

An important problem in knowledge discovery from text is the automatic extraction of semantic relations. This paper presents a supervised, semantically intensive, domain independent approach for the automatic detection of part-whole relations in text. First an algorithm is described that identifies lexico-syntactic patterns that encode part-whole relations. A difficulty is that these patterns also encode other semantic relations, and a learning method is necessary to discriminate whether or not a pattern contains a part-whole relation. A large set of training examples have been annotated and fed into a specialized learning system that learns classification rules. The rules are learned through an iterative semantic specialization (ISS) method applied to noun phrase constituents. Classification rules have been generated this way for different patterns such as genitives, noun compounds, and noun phrases containing prepositional phrases to extract part-whole relations from them. The applicability of these rules has been tested on a test corpus obtaining an overall average precision of 80.95% and recall of 75.91%. The results demonstrate the importance of word sense disambiguation for this task. They also demonstrate that different lexico-syntactic patterns encode different semantic information and should be treated separately in the sense that different clarification rules apply to different patterns.

We present a data-driven approach using word embeddings to discover and categorise language biases on the discussion platform Reddit. As spaces for isolated user communities, platforms such as Reddit are increasingly connected to issues of racism, sexism and other forms of discrimination, signalling the need to monitor the language of these groups. One of the most promising AI approaches to trace linguistic biases in large textual datasets involves word embeddings, which transform text into high-dimensional dense vectors and capture semantic relations between words. Yet, previous studies require predefined sets of potential biases to study, e.g., whether gender is more or less associated with particular types of jobs. This makes these approaches unfit to deal with smaller and community-centric datasets such as those on Reddit, which contain smaller vocabularies and slang, as well as biases that may be particular to that community. This paper proposes a data-driven approach to automatically discover language biases encoded in the vocabulary of online discourse communities on Reddit. In our approach, protected attributes are connected to evaluative words found in the data, which are then categorised through a semantic analysis system. We verify the effectiveness of our method by comparing the biases we discover in the Google News dataset with those found in previous literature. We then successfully discover gender bias, religion bias, and ethnic bias in different Reddit communities. We conclude by discussing potential application scenarios and limitations of this data-driven bias discovery method.

The discovery of words in continuous speech is one of the first challenges faced by infants during language acquisition. This process is partially facilitated by statistical learning, the ability to discover and encode relevant patterns in the environment. Here, we used an electroencephalogram (EEG) index of neural entrainment to track 6-month-olds’ ( N = 25) segmentation of words from continuous speech. Infants’ neural entrainment to embedded words increased logarithmically over the learning period, consistent with a perceptual shift from isolated syllables to wordlike units. Moreover, infants’ neural entrainment during learning predicted postlearning behavioral measures of word discrimination ( n = 18). Finally, the logarithmic increase in entrainment to words was comparable in infants and adults, suggesting that infants and adults follow similar learning trajectories when tracking probability information among speech sounds. Statistical-learning effects in infants and adults may reflect overlapping neural mechanisms, which emerge early in life and are maintained throughout the life span.

Abstract Abstract 685 Allogeneic stem cell transplantation (allo-SCT) followed by donor lymphocyte infusion (DLI) is used as a curative treatment for patients with malignant diseases. Donor derived T cells mediate graft versus tumor responses by targeting minor histocompatibility antigens (mHags) that are encoded by patient specific single nucleotide polymorphisms (SNPs). Various approaches have been applied for mHag discovery resulting in the characterization of more than 20 mHags. However, restriction to specific HLA types and unfavourable gene expression strongly limits the number of clinically relevant mHags. Recently, it has been demonstrated that whole genome association scanning (WGAs) can be a tool for mHag identification. Here, we present WGAs as a powerful method for high throughput identification of new mHags. From 2 patients that entered complete remission with limited graft versus host disease after allo-SCT and DLI, activated T cells were cloned by flowcytometric cell sorting. After expansion, 232 stably growing T cell clones were obtained. Patient specific recognition in the absence of donor recognition was demonstrated for 78 clones. By using blocking antibodies and a test panel consisting of partially HLA matched EBV-transformed B cell lines (EBV-LCL), we demonstrated that these 78 T cell clones comprised 20 unique mHag reactivities which could be identified to be restricted to HLA-A*02 or B*07. Since WGAs is based on a balanced segregation of test cells in a positive and a negative group, 15 T cell clones were selected recognizing mHags with population frequencies between 20% to 80% for further analysis. To perform WGAs, a test panel was generated containing 80 HLA-A*02 and B*07 positive EBV-LCL for testing of recognition by all selected T cell clones using Interferon-γ Elisa. In parallel, all EBV-LCL were genotyped for 1 million SNPs using bead arrays. All SNP genotype patterns were combined with each individual T cell recognition pattern. The level of matching between both patterns was statistically analyzed using Fisher's exact test, resulting in p-values indicating the significance of association. Significant association (p-value<10-12) between SNP genotypes and a T cell recognition pattern identified a single genomic region for 12 out of 15 T cell clones. In 2 cases no clear discrimination between positive and negative EBV-LCL could be made, suggesting that these T cell clones may not recognize mHags. Incomplete coverage of a genomic region by SNPs on the bead array may explain the lack of association for 1 T cell clone. For 7 T cell clones, significant association was found with array SNPs located in exons of the genes WNK1, SSR1, PRCP, ARHGDIB, PDCD11, EBI3 and APOBEC3B. For 3 other T cell clones, the genes ERAP1, BCAT2 and GEMIN4 were identified based on significant association with SNPs located in non coding regions. Sequence analysis of the coding regions of these genes from patient and donor revealed additional patient specific SNPs that were not included in the bead array. For the remaining associating TTK and ERGIC1 genes, the coding regions were identical between patient and donor, showing that these mHags are not encoded by exon SNPs in the identified TTK and ERGIC1 genes. Differential mHag expression may be induced in these cases by SNPs in adjacent genes that were not identified by SNPs on the bead array or may be the result of SNPs in non coding regulatory regions or in mRNA splice variants. According to the BioGPS gene expression database, a number of genes as identified by WGAs were predominantly expressed in hematopoietic cells, and may therefore encode relevant targets for T cell therapy. Next, we investigated the amino acid polymorphisms encoded by all identified coding SNPs. Peptides spanning the patient type amino acid polymorphism were submitted to HLA binding prediction algorithms. Candidate peptides were synthesized and T cell recognition was demonstrated at concentrations varying from 0.5 to 5000 nM. Recognition of donor type peptides was absent in all cases, validating the identification of 10 novel mHags. In conclusion, these data demonstrate that activation marker based T cell selection and cloning combined with WGAs results in high throughput discovery of multiple mHags. This strategy therefore allows broad characterization of mHags in donor derived T cell responses and selection of clinically relevant mHags for development of T cell based immunotherapy. Disclosures: No relevant conflicts of interest to declare.

Person re-identification (ReID) plays an important role in intelligent surveillance and receives widespread attention from academics and the industry. Due to extreme changes in viewing angles, some discriminative local regions are suppressed. In addition, the data with similar backgrounds collected by a fixed viewing angle camera will also affect the model’s ability to distinguish a person. Therefore, we need to discover more fine-grained information to form the overall characteristics of each identity. The proposed self-erasing network structure composed of three branches benefits the extraction of global information, the suppression of background noise and the mining of local information. The two self-erasing strategies that we proposed encourage the network to focus on foreground information and strengthen the model’s ability to encode weak features so as to form more effective and richer visual cues of a person. Extensive experiments show that the proposed method is competitive with the advanced methods and achieves state-of-the-art performance on DukeMTMC-ReID and CUHK-03(D) datasets. Furthermore, it can be seen from the activation map that the proposed method is beneficial to spread the attention to the whole body. Both metrics and the activation map validate the effectiveness of our proposed method.

483 Background: Cystoscopy, an invasive, painful and expensive method, is currently the main clinical tool for new diagnosis and disease recurrence detection of bladder cancer (BCa). The need for new biomarkers discovery that is costless, sensitive and specific is urgent. This project aims to study and compare alternative splicing events (ASE) in BCa tissues and normal bladder tissues and ultimately, identify specific spliced events coding for proteins detectable in urine by liquid chromatography–mass spectrometry. Methods: In this study, alterations to the global RNA splicing landscape of cellular genes were investigated in a large-scale screen from 408 BCa tissues and 19 normal tissues provided by The Cancer Genome Atlas (TCGA). Three statistical thresholds were used to determine substantial modifications. All events showing a p-value<0.05 and a level of expression ≥ 50 transcripts per million; -10 ≥ Δ percent splice index ≤10; and a q-value<0.05 were conserved. Next, mRNA expression levels between cancer and normal tissues were compared for all splicing factors and the spliceosome to determine the impact of gene dysregulation on alternative splicing events. Using multiple bioinformatic platforms such as EASANA, MultAlin, ExPasy, NLS Mapper and Pfam, splicing events responsible for significant protein structural changes between cancer and healthy tissue were selected. From this sample chosen, ASEs coding for proteins that could be detected in urine were conserved. Results: Our study identifies modifications in the alternative splicing patterns of 107 transcripts encoded by 97 genes. STRING analysis revealed that many of the gene products interact either directly or indirectly with each other (enrichment p-value = 1x10-10). 61 ASEs are causing important protein changes from which 27 can be detected in urine. Finally, 16 ASEs coding for easily recognizable peptide sequences in urine represented significant targets for potential BCa biomarkers. Conclusions: The TCGA data show the relevance to investigate alternative splicing events in bladder cancer. 16 significant events were detectable in urine and may potentially discriminate between presence or absence of bladder cancer.

Much of the research on gender representations in language teaching materials has focused on providing frequency-based accounts of character appearances, familial/occupational role attributions and sexist language use. However, gender discrimination, when communicated visually, might more readily drop off the already overburdened teacher’s radar. Therefore, this study concentrated not just on the depiction of coursebook images but also on their relation to the learners, and aimed to discover the latent sexism in three thematically similar units from one global and two locally-produced coursebooks widely used in the Turkish and Iranian EFL contexts. A critical analysis of 41 images with Van Leeuwen’s (2008) framework revealed that male overrepresentation prevailed throughout all three resources, though to a lesser extent in the global coursebook. The characters mainly avoided direct contact with the viewers by averting their gaze and offered themselves as visual cues for denotative meanings. The global and Turkish-made series tended to position them both closer to the young readers and at their eye level to help build intimacy with more relatable role models. In their Iranian counterpart, the male and female characters were yet socially distanced from them through long shots taken from low and high angles respectively, in which case men were portrayed as authority figures to be looked up to, and women as the diminished other to be looked down on by the students. While both genders were oftener seen frontally in the Turkish EFL material with mixed-gender authorship, the all-male author teams preferred to show the male characters from an oblique angle to further detachment in the global and Iranian contexts. In establishing relatively closer, more personal and engaging interactions with both boys and girls visually, the global and Turkish EFL materials can be claimed to encode a more inclusive and equitable worldview than their Iranian counterpart.

Abstract In recent years, deep neural networks have achieved extraordinary performance on supervised learning tasks. Convolutional neural networks (CNN) have vastly improved the state of the art for most computer vision tasks including object recognition and segmentation. However, their success relies on the presence of a large amount of labeled data. In contrast, relatively fewer work has been done in deep learning to handle scenarios when access to ground truth is limited, partial or completely absent. In this thesis, we propose models to handle challenging problems with limited labeled information. Our first contribution is a neural architecture that allows for the extraction of infinitely many features from an object while allowing for tractable inference. This is achieved by using the `kernel trick', that is, we express the inner product in the infinite dimensional feature space as a kernel. The kernel can either be computed exactly for single layer feedforward networks, or approximated by an iterative algorithm for deep convolutional networks. The corresponding models are referred to as stretched deep networks (SDN). We show that when the amount of training data is limited, SDNs with random weights drastically outperform fully supervised CNNs with similar architectures. While SDNs perform reasonably well for classification with limited labeled data, they can not utilize unlabeled data which is often much easier to obtain. A common approach to utilize unlabeled data is to couple the classifier with an autoencoder (or its variants) thereby minimizing reconstruction error in addition to the classification error. We discuss the limitations of decoder based architectures and propose a model that allows for the utilization of unlabeled data without the need of a decoder. This is achieved by jointly modeling the distribution of data and latent features in a manner that explicitly assigns zero probability to unobserved data. The joint probability of the data and the latent features is maximized using a two-step EM-like procedure. Depending on the task, we allow the latent features to be one-hot or real-valued vectors and define a suitable prior on the features. For instance, one-hot features correspond to class labels and are directly used for the unsupervised and semi-supervised classification tasks. For real-valued features, we use hierarchical Bayesian models as priors over the latent features. Hence, the proposed model, which we refer to as discriminative encoder (or DisCoder), is flexible in the type of latent features that it can capture. The proposed model achieves state-of-the-art performance on several challenging datasets. Having addressed the problem of utilizing unlabeled data for classification, we move to a domain where obtaining labels is a lot more expensive, that is, semantic segmentation of images. Explicitly labeling each pixel of an image with the object that the pixel belongs to, is an expensive operation, in terms of time as well as effort? Currently, only a few classes of images have been densely (pixel-level) labeled. Even among these classes, only a few images per class have pixel-level supervision. Models that rely on densely-labeled images, cannot utilize a much larger set of weakly annotated images available on the web. Moreover, these models cannot learn the segmentation masks for new classes, where there is no densely labeled data. Hence, we propose a model for utilizing weakly-labeled data for semantic segmentation of images. This is achieved by generating fake labels for each image, while simultaneously forcing the output of the CNN to satisfy the mean-field constraints imposed by a conditional random field. We show that one can enforce the CRF constraints by forcing the distribution at each pixel to be close to the distribution of its neighbors. The proposed model is very fast to train and achieves state-of-the-art performance on the popular VOC-2012 dataset for the task of weakly supervised semantic segmentation of images.

Cross-modal hashing intends to project data from two modalities into a common hamming space to perform cross-modal retrieval efficiently. Despite satisfactory performance achieved on real applications, existing methods are incapable of effectively preserving semantic structure to maintain inter-class relationship and improving discriminability to make intra-class samples aggregated simultaneously, which thus limits the higher retrieval performance. To handle this problem, we propose Equally-Guided Discriminative Hashing (EGDH), which jointly takes into consideration semantic structure and discriminability. Specifically, we discover the connection between semantic structure preserving and discriminative methods. Based on it, we directly encode multi-label annotations that act as high-level semantic features to build a common semantic structure preserving classifier. With the common classifier to guide the learning of different modal hash functions equally, hash codes of samples are intra-class aggregated and inter-class relationship preserving. Experimental results on two benchmark datasets demonstrate the superiority of EGDH compared with the state-of-the-arts.

Current weakly-supervised semantic segmentation (WSSS) methods with image-level labels mainly adopt class activation maps (CAM) to generate the initial pseudo labels. However, CAM usually only identifies the most discriminative object extents, which is attributed to the fact that the network doesn't need to discover the integral object to recognize image-level labels. In this work, to tackle this problem, we proposed to simultaneously learn the image-level labels and local visual word labels. Specifically, in each forward propagation, the feature maps of the input image will be encoded to visual words with a learnable codebook. By enforcing the network to classify the encoded fine-grained visual words, the generated CAM could cover more semantic regions. Besides, we also proposed a hybrid spatial pyramid pooling module that could preserve local maximum and global average values of feature maps, so that more object details and less background were considered. Based on the proposed methods, we conducted experiments on the PASCAL VOC 2012 dataset. Our proposed method achieved 67.2% mIoU on the val set and 67.3% mIoU on the test set, which outperformed recent state-of-the-art methods.