Fig. 4From: Multiple instance neural networks based on sparse attention for cancer detection using T-cell receptor sequencesHeatmap of estimated attention weights via the sparsemax (left) and softmax (right) function. Samples are reordered by the number of instances and the masking area (no instances) is colored in gray. Red and green labels indicate tumor and normal samples, respectively. Here, we show the balanced case for DLBC cancer dataBack to article page