top of page

Craft, activity and play ideas


Deep Red 128x

As we have worked with dozens of AMD EPYC platforms and published many reviews, a common question is why the platforms do not expose 128x PCIe lanes to the motherboard or server in the AMD EPYC 7001 generation. Here is a good example of where that is in play on a Gigabyte G291-Z20 dual GPU EPYC platform.

Deep Red 128x

If you had to pick one deep learning technique for computer vision from the plethora of options out there, which one would you go for? For a lot of folks, including myself, convolutional neural network is the default answer.

The previous articles of this series covered the basics of deep learning and neural networks. We also learned how to improve the performance of a deep neural network using techniques like hyperparameter tuning, regularization and optimization.

There are a number of hyperparameters that we can tweak while building a convolutional network. These include the number of filters, size of filters, stride to be used, padding, etc. We will look at each of these in detail later in this article. Just keep in mind that as we go deeper into the network, the size of the image shrinks whereas the number of channels usually increases.

Generally, we take the set of hyperparameters which have been used in proven research and they end up doing well. As seen in the above example, the height and width of the input shrinks as we go deeper into the network (from 32 X 32 to 5 X 5) and the number of channels increases (from 3 to 10).

Training very deep networks can lead to problems like vanishing and exploding gradients. How do we deal with these issues? We can use skip connections where we take activations from one layer and feed it to another layer that is even more deeper in the network. There are residual blocks in ResNet which help in training deeper networks.

The benefit of training a residual network is that even if we train deeper networks, the training error does not increase. Whereas in case of a plain network, the training error first decreases as we train a deeper network and then starts to rapidly increase:

Suppose we use the lth layer to define the content cost function of a neural style transfer algorithm. Generally, the layer which is neither too shallow nor too deep is chosen as the lth layer for the content cost function. We use a pretrained ConvNet and take the activations of its lth layer for both the content image as well as the generated image and compare how similar their content is. With me so far?

Recent technology advances, in which histological tumor slides are converted to digital image datasets and in machine learning which can interrogate patterns in digital images of MRI and digitized histology may address some of these limitations. Specifically, for response evaluation in osteosarcoma, digital histopathology is made possible by scanning hematoxylin and eosin (H&E) stained microscopic slides [14] using commercially available scanning technology. The scanning converts a glass slide to a digital Whole Slide Image (WSI) that preserves image resolution up to 40X magnification. Although each WSI represents a very large digital file, interpretation is now feasible using image processing algorithms, while the advent of machine and deep-learning models make automated diagnostic systems a possibility.

Step 1 shows the assembly of patient archival samples of 50 cases, resulting in 942 WSIs. Step 2 involves the selection of 40 handpicked WSIs by a pathologist. In step 3, 1144 image tiles of size 1024x1024 are generated from WSIs identified in step 2. From each image tile in step 3, a number of image patches of size 128x128 are generated.

After retraining, the performance of SVM3 and the deep-learner in discrimination of non-tumor from tumor, followed by conditional discrimination of necrotic from viable tumor, was further assessed on the test data set by constructing the corresponding hierarchical receiver-operator characteristics. Receiver operating characteristic (ROC) curve analysis is widely used in biomedical research to assess the performance of diagnostic tests. The ROC curve depicts the quality of a diagnostic marker in a two-class classification problem. It illustrates the trade-off between sensitivity and specificity as a cut-off point for decision making. Area Under the ROC Curve (AUC) is the most widely used index for the quantification of the performance of a diagnostic marker in the two-class setting. Fig 4 shows the ROC curves for Tumor vs. Non-Tumor for the SVM classifier and deep-learner. For a 3-class setting, the Volume Under the Surface (VUS) was proposed as an index for the assessment of the diagnostic accuracy of the marker under consideration [35]. These surfaces are shown in Fig 5.

Receiver-operating characteristic surface for discrimination of non-tumor from tumor followed by conditional discrimination of necrotic from viable tumor with volume under surface (VUS) for (a) SVM3, (b) deep-learner. True positive rates (TPR) are within-class fractions of correctly classified image tiles/patches.

To our knowledge, the data presented in this report represent the first description using automated learner tools in the histological classification in high-grade osteosarcoma. We have optimized a pipeline for this interpretation, and highlighted multiple novel achievements toward this end. We developed an annotation tool for expert pathology review of osteosarcoma tiles for classification purpose, which allowed efficient and expert review of 1,144 image tiles, while simultaneously generating data for our machine-learning and deep-learning algorithms. We have optimized an image-processing platform using the Cellprofiler software, and identified 53 features in addition to those distilled and extrapolated from expert pathology examination. The high VUS value for both learners prove the robustness of the learned models on the blind test dataset.

Researchers from the University of Illinois have developed a technology that enables ultra high bandwidth data communication as well as power transfer over short distances using acoustic signal.The technology works when items are not touching, it works without pins that could bend or break, it works when moving around and only "near' connected. This technology can be used in any media which allows acoustic waves such as underwater in deep ocean, human-tissue, etc. to achieve data rates in excess of 300Mbps. 041b061a72


Welcome to the group! You can connect with other members, ge...
bottom of page