REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasets

International Journal of Computer Vision - Machine learning models are known to perpetuate and even amplify the biases present in the data. However, these data biases frequently do not become...     

Alpha-melanocyte stimulating hormone-induced pigmentation is blocked by depletion of protein kinase C

Partial regions of the mRNA encoding a major part of translation elongation factor 2 (EF-2) from a kinetoplastid protozoan, Trypanosoma cruzi, were amplified by means of polymerase chain reaction and their primary structures were analyzed. The deduced amino acid sequence was aligned with those of other eukaryotic and archaebacterial EF-2s, and the phylogenetic relationships among eukaryotes were inferred by the maximum likelihood (ML) method. ML analyses of EF-2 phylogeny using six different stochastic models of amino acid substitutions consistently suggested that the phylogenetic position of T. cruzi is likely to be closer to higher eukaryotes than that inferred from the phylogeny of small subunit ribosomal RNA (SrRNA). These results are consistent with those for the elongation factor 1alpha (EF-1alpha) phylogeny. When the EF-1alpha and EF-2 phylogenies were totally evaluated, it became much clearer that the divergence of T. cruzi occurred later than that of a mitochondrion-lacking protozoan, Entamoeba histolytica, although this is not conclusive.     

Every Moment Counts: Dense Detailed Labeling of Actions in Complex Videos

Every moment counts in action recognition. A comprehensive understanding of human activity in video requires labeling every frame according to the actions occurring, placing multiple labels densely over a video sequence. To study this problem we extend the existing THUMOS dataset and introduce MultiTHUMOS, a new dataset of dense labels over unconstrained internet videos. Modeling multiple, dense labels benefits from temporal relations within and across classes. We define a novel variant of long short-term memory deep networks for modeling these temporal relations via multiple input and output connections. We show that this model improves action labeling accuracy and further enables deeper understanding tasks ranging from structured retrieval to action prediction.