Deep Project Research Articles

DeepProjection: specific and robust projection of curved 2D tissue sheets from 3D microscopy using deep learning.

The efficient extraction of image data from curved tissue sheets embedded in volumetric imaging data remains a serious and unsolved problem in quantitative studies of embryogenesis. Here, we present DeepProjection (DP), a trainable projection algorithm based on deep learning. This algorithm is trained on user-generated training data to locally classify 3D stack content, and to rapidly and robustly predict binary masks containing the target content, e.g. tissue boundaries, while masking highly fluorescent out-of-plane artifacts. A projection of the masked 3D stack then yields background-free 2D images with undistorted fluorescence intensity values. The binary masks can further be applied to other fluorescent channels or to extract local tissue curvature. DP is designed as a first processing step than can be followed, for example, by segmentation to track cell fate. We apply DP to follow the dynamic movements of 2D-tissue sheets during dorsal closure in Drosophila embryos and of the periderm layer in the elongating Danio embryo. DeepProjection is available as a fully documented Python package.

Adaptive gradients and weight projection based on quantized neural networks for efficient image classification

Quantization of weights and activations has been introduced into optimization methodologies of deep neural networks (DNNs) to address issues like memory consumption and massive computational demands that impede edge applications of neural networks. There are multiple approaches to quantization which can probably all be classified into two categories: uniform quantization and non-uniform quantization. The key difference between these methods is whether unequal quantization intervals are used to match the non-uniform distribution of weights. However, conventional techniques train weights layer by layer to suit previously determined quantization points, which poses difficulty for reaching optimal points. We proposed a framework called Deep Projection (DP) to train quantized weights with adaptive gradients. Gradients and weights are projected to high dimensional training space through projection layers, resulting in more complicated update paths and randomness, which benefit the generalization of models. Renewal of weights in the network is conducted by values composited by self-updating high dimensional tensors, which means that the resulting gradients are softer and more adaptive to network training. We applied our method via a uniform quantization approach, and the results showed improvements even when we limited the first and last layers with lower bit-width. The results could be comparable with the non-uniform quantization method with 4-bit precision. Distributions of weights in different quantized networks are analyzed to display the advantages of our method. Besides, all of these steps can be realized in a conventional training process. There are no limitations to quantization bit-width and function. Namely, our framework is readily implemented with a concise architecture.

The Trap of Abstract Space: Recomposing Russian-Speaking Immigrants in Post-Soviet Estonia

Focusing on the Estonian government's policy to integrate Soviet-era Russian-speaking immigrants, this article historically examines how public officials have come to constitute Estonian state space as a particular national place through the re-constitution of these immigrants in public policy. Drawing on Lefebvre's The Production of Space , the analysis describes place as a combined result of three types of space mobilized in social practice: spatial practices, representational spaces, and representations of space. Conceived through the third type, successful statecraft produces an abstract space that traps target populations through subject-making. The article's deeper project is two-fold. First, it aims to theorize the spatial interplay between deterritorialization and the state's efforts to reterritorialize the nation through citizenship and integration policy. Second, it aims to compensate for ethnographic tendencies that privilege data obtained through direct sensory experience to better capture how the deployment of spatial abstractions in public policy marginalize immigrants and minority groups.