Bharath Hariharan

Microtubules (MTs) perform important cellular functions including migration, intracellular trafficking, and chromosome segregation. The centrosome, comprised of two centrioles surrounded by the pericentriolar material (PCM), is the cell’s central MT organizing center. The PCM proteins, includingγ-tubulin and Pericentrin, promote MT nucleation and organization. Centrosomes in cancer cells are commonly numerically amplified. However, the question of how amplification of centrosomes alters the MT organization capacity is not well-studied. We developed a quantitative image-processing and machine learning-aided approach for the automated analysis of MT organization. We designed a convolutional neural network-based approach for detecting centrosomes and an automated pipeline for analyzing MT organization around centrosomes, encapsulated in a semi-automatic graphical tool. Using this tool, we analyzed the spatial distribution of PCM proteins, the growing ends of MTs and the total MT density in breast cancer cells. We find that breast cancer cells with supernumerary centrosomes not only have increased PCM protein but also exhibit expansion in PCM size. Moreover, centrosome amplified cells have a greater MT density and more growing MT ends near centrosomes than unamplified cells. The semi-automated approach developed here enables facile, unbiased and quantitative measurements of centrosome aberrations. We show that these aberrations increase MT nucleation and promote changes to MT density and the spatial distribution of MTs around amplified centrosomes.

In this paper, we present a thorough evaluation of the efficacy of knowledge distillation and its dependence on student and teacher architectures. Starting with the observation that more accurate teachers often don't make good teachers, we attempt to tease apart the factors that affect knowledge distillation performance. We find crucially that larger models do not often make better teachers. We show that this is a consequence of mismatched capacity, and that small students are unable to mimic large teachers. We find typical ways of circumventing this (such as performing a sequence of knowledge distillation steps) to be ineffective. Finally, we show that this effect can be mitigated by stopping the teacher's training early. Our results generalize across datasets and models.

Traditional recognition methods typically require large, artificially-balanced training classes, while few-shot learning methods are tested on artificially small ones. In contrast to both extremes, real world recognition problems exhibit heavy-tailed class distributions, with cluttered scenes and a mix of coarse and fine-grained class distinctions. We show that prior methods designed for few-shot learning do not work out of the box in these challenging conditions, based on a new "meta-iNat" benchmark. We introduce three parameter-free improvements: (a) better training procedures based on adapting cross-validation to meta-learning, (b) novel architectures that localize objects using limited bounding box annotations before classification, and (c) simple parameter-free expansions of the feature space based on bilinear pooling. Together, these improvements double the accuracy of state-of-the-art models on meta-iNat while generalizing to prior benchmarks, complex neural architectures, and settings with substantial domain shift.