RESUMO
How to accurately learn task-relevant state representations from high-dimensional observations with visual distractions is a realistic and challenging problem in visual reinforcement learning. Recently, unsupervised representation learning methods based on bisimulation metrics, contrast, prediction, and reconstruction have shown the ability for task-relevant information extraction. However, due to the lack of appropriate mechanisms for the extraction of task information in the prediction, contrast, and reconstruction-related approaches and the limitations of bisimulation-related methods in domains with sparse rewards, it is still difficult for these methods to be effectively extended to environments with distractions. To alleviate these problems, in the paper, the action sequences, which contain task-intensive signals, are incorporated into representation learning. Specifically, we propose a Sequential Action-induced invariant Representation (SAR) method, which decouples the controlled part (i.e., task-relevant information) and the uncontrolled part (i.e., task-irrelevant information) in noisy observations through sequential actions, thereby extracting effective representations related to decision tasks. To achieve it, the characteristic function of the action sequence's probability distribution is modeled to specifically optimize the state encoder. We conduct extensive experiments on the distracting DeepMind Control suite while achieving the best performance over strong baselines. We also demonstrate the effectiveness of our method at disregarding task-irrelevant information by applying SAR to real-world CARLA-based autonomous driving with natural distractions. Finally, we provide the analysis results of generalization drawn from the generalization decay and t-SNE visualization. Code and demo videos are available at https://github.com/DMU-XMU/SAR.git.
RESUMO
Tissue-based gene expression data analyses, while most powerful, represent a significantly more challenging problem compared to cell-based gene expression data analyses, even for the simplest differential gene expression analyses. The result in determining if a gene is differentially expressed in tumor vs. non-tumorous control tissues does not only depend on the two expression values but also on the percentage of the tissue cells being tumor cells, i.e., the tumor purity. We developed a novel matched-pairs feature selection method, which takes into full consideration of the tumor purity when deciding if a gene is differentially expressed in tumor vs. control experiments, which is simple, effective, and accurate. To evaluate the validity and performance of the method, we have compared it with four published methods using both simulated datasets and actual cancer tissue datasets and found that our method achieved better performance with higher sensitivity and specificity than the other methods. Our method was the a matched-pairs feature selection method on gene expression analysis under matched case-control design which takes into consideration the tumor purity information, which can set a foundation for further development of other gene expression analysis needs.
Assuntos
Perfilação da Expressão Gênica , Modelos Biológicos , Modelos Estatísticos , Neoplasias/genética , Neoplasias/patologia , Projetos de Pesquisa/normas , Estudos de Casos e Controles , Humanos , Sensibilidade e EspecificidadeRESUMO
Non-invasive prediction of isocitrate dehydrogenase (IDH) genotype plays an important role in tumor glioma diagnosis and prognosis. Recently, research has shown that radiology images can be a potential tool for genotype prediction, and fusion of multi-modality data by deep learning methods can further provide complementary information to enhance prediction accuracy. However, it still does not have an effective deep learning architecture to predict IDH genotype with three-dimensional (3D) multimodal medical images. In this paper, we proposed a novel multimodal 3D DenseNet (M3D-DenseNet) model to predict IDH genotypes with multimodal magnetic resonance imaging (MRI) data. To evaluate its performance, we conducted experiments on the BRATS-2017 and The Cancer Genome Atlas breast invasive carcinoma (TCGA-BRCA) dataset to get image data as input and gene mutation information as the target, respectively. We achieved 84.6% accuracy (area under the curve (AUC) = 85.7%) on the validation dataset. To evaluate its generalizability, we applied transfer learning techniques to predict World Health Organization (WHO) grade status, which also achieved a high accuracy of 91.4% (AUC = 94.8%) on validation dataset. With the properties of automatic feature extraction, and effective and high generalizability, M3D-DenseNet can serve as a useful method for other multimodal radiogenomics problems and has the potential to be applied in clinical decision making.
RESUMO
A soft-modeling multivariate numerical approach that combines self-modeling curve resolution (SMCR) and mixed Lorentzian-Gaussian curve fitting was successfully implemented for the first time to elucidate spatially and spectroscopically resolved spectral information from infrared imaging data of oral mucosa cells. A novel variant form of the robust band-target entropy minimization (BTEM) SMCR technique, coined as hierarchical BTEM (hBTEM), was introduced to first cluster similar cellular infrared spectra using the unsupervised hierarchical leader-follower cluster analysis (LFCA) and subsequently apply BTEM to clustered subsets of data to reconstruct three protein secondary structure (PSS) pure component spectra-alpha-helix, beta-sheet, and ambiguous structures-that associate with spatially differentiated regions of the cell infrared image. The Pearson VII curve-fitting procedure, which approximates a mixed Lorentzian-Gaussian model for spectral band shape, was used to optimally curve fit the resolved amide I and II bands of various hBTEM reconstructed PSS pure component spectra. The optimized Pearson VII band-shape parameters and peak center positions serve as means to characterize amide bands of PSS spectra found in various cell locations and for approximating their actual amide I/II intensity ratios. The new hBTEM methodology can also be potentially applied to vibrational spectroscopic datasets with dynamic or spatial variations arising from chemical reactions, physical perturbations, pathological states, and the like.