Pytorch installation on Windows is a pain and Tensorflow isn't available on Python 2.7 for windows which ensues in a nice segue to the solution… You can use this blog post either as a reference guide to reinstall your bash and expiate changing permissions for ssh host keys and messing with chmod on the command line like I did,or as a starting point to put aside dogma and try something new and interesting. Going on a wild goose chase to reinstall modules or packages and restore everything by making a myriad of setting changes, gets me super nettled. It's cumbersome and tedious, so to save my time and sanity, lest it should happen in the future again, I've gathered a cornucopia of commands and install guidelines to ensure a clean and successful workspace for Machine Learning and Deep Learning code. A good IDE is conducive to efficient and effective coding practices. One of the best IDE for all things Python related, that I have come across, is Pycharm.

My IBM Research AI team and I recently completed the first formal theoretical study of the convergence rate and communications complexity associated with a decentralized distributed approach in a deep learning training setting. The empirical evidence proves that in specific configurations, a decentralized approach can result in a 10x performance boost over a centralized approach without additional complexity. A paper describing our work has been accepted for oral presentation at the NIPS 2017 Conference, one of the 40 out of 3240 submissions selected for this. Supervised machine learning generally consists of two phases: 1) training (building a model) and 2) inference (making predictions with the model). The training phase involves finding optimal values for a model's parameters such that error on a set of training examples is minimized, and the model generalizes to new data.

Deep learning holds many mysteries for theory, as we have discussed on this blog. Lately many ML theorists have become interested in the generalization mystery: why do trained deep nets perform well on previously unseen data, even though they have way more free parameters than the number of datapoints (the classic "overfitting" regime)? Zhang et al.'s paper Understanding Deep Learning requires Rethinking Generalization played some role in bringing attention to this challenge. Their main experimental finding is that if you take a classic convnet architecture, say Alexnet, and train it on images with random labels, then you can still achieve very high accuracy on the training data. Needless to say, the trained net is subsequently unable to predict the (random) labels of still-unseen images, which means it doesn't generalize.

A revolution in AI is occurring thanks to progress in deep learning. How far are we towards the goal of achieving human-level AI? What are some of the main challenges ahead? Yoshua Bengio believes that understanding the basics of AI is within every citizen's reach. That democratizing these issues is important so that our societies can make the best collective decisions regarding the major changes AI will bring, thus making these changes beneficial and advantageous for all.

Several weeks ago, Jefferies' analyst James Kisner published a scathing report, shedding light onto the shortcomings of IBM Watson. Kisner focused on the $60 million disastrous Watson project for MD Anderson, and highlighted how much IBM is lagging behind Amazon and Apple. As John Mannes pointed out on TechCrunch, "things would look much worse if Google, Microsoft and Facebook were added to this table." He also eloquently summarized the common pitfall in our approach to AI: "Reality is that AI isn't an amorphous black hole that sucks in unstructured data to produce insights. A solid data pipeline and a domain-specific understanding of the AI business problem at hand is table minimum."

Computer algorithms analyzing digital pathology slide images were shown to detect the spread of cancer to lymph nodes in women with breast cancer as well as or better than pathologists, in a new study published online in the Journal of the American Medical Association.1 Researchers competed in an international challenge in 2016 to produce computer algorithms to detect the spread of breast cancer by analyzing tissue slides of sentinel lymph nodes, the lymph node closest to a tumor and the first place it would spread. The performance of the algorithms was compared against the performance of a panel of pathologists participating in a simulation exercise. Images of lymph node tissue sections used to test the ability of the deep learning algorithms to detect cancer metastasis. Specifically, in cross-sectional analyses that evaluated 32 algorithms, seven deep learning algorithms showed greater discrimination than a panel of 11 pathologists in a simulated time-constrained diagnostic setting, with an area under the curve of 0.994 (best algorithm) versus 0.884 (best pathologist). The study found that some computer algorithms were better at detecting cancer spread than pathologists in an exercise that mimicked routine pathology workflow.

AWS DeepLens Looking for a new way to learn machine learning? Let a machine teach you with AWS DeepLens, the world's first deep learning enabled video camera for developers. Designed to connect securely to a variety of AWS offerings, including AWS IoT, Amazon SQS, Amazon SNS, and Amazon DynamoDB, AWS DeepLens uses Amazon Kinesis Video Streams to stream video back to AWS and Amazon Rekognition Video to apply advanced video analytics. Easy to customize and fully programmable with AWS Lambda, AWS DeepLens runs on any deep learning framework, including TensorFlow and Caffe.

The most challenging part of deep learning is labeling, as you'll see in part one of this two-part series, Learn how to classify images with TensorFlow. Proper training is critical to effective future classification, and for training to work, we need lots of accurately labeled data. In part one, I skipped over this challenge by downloading 3,000 prelabeled images. I then showed you how to use this labeled data to train your classifier with TensorFlow.

The flurry of headlines surrounding AlphaGo Zero (the most recent version of DeepMind's AI system for playing Go) means interest in reinforcement learning (RL) is bound to increase. Next to deep learning, RL is among the most followed topics in AI. For most companies, RL is something to investigate and evaluate but few organizations have identified use cases where RL may play a role. As we enter 2018, I want to briefly describe areas where RL has been applied. RL is confusingly used to refer to a set of problems and a set of techniques, so let's first settle on what RL will mean for the rest of this post.

Objective: Predict patient-specific vitals deemed medically acceptable for discharge from a pediatric intensive care unit (ICU). Design: The means of each patient's hr, sbp and dbp measurements between their medical and physical discharge from the ICU were computed as a proxy for their physiologically acceptable state space (PASS) for successful ICU discharge. These individual PASS values were compared via root mean squared error (rMSE) to population age-normal vitals, a polynomial regression through the PASS values of a Pediatric ICU (PICU) population and predictions from two recurrent neural network models designed to predict personalized PASS within the first twelve hours following ICU admission. Setting: PICU at Children's Hospital Los Angeles (CHLA). Patients: 6,899 PICU episodes (5,464 patients) collected between 2009 and 2016. Interventions: None. Measurements: Each episode data contained 375 variables representing vitals, labs, interventions, and drugs. They also included a time indicator for PICU medical discharge and physical discharge. Main Results: The rMSEs between individual PASS values and population age-normals (hr: 25.9 bpm, sbp: 13.4 mmHg, dbp: 13.0 mmHg) were larger than the rMSEs corresponding to the polynomial regression (hr: 19.1 bpm, sbp: 12.3 mmHg, dbp: 10.8 mmHg). The rMSEs from the best performing RNN model were the lowest (hr: 16.4 bpm; sbp: 9.9 mmHg, dbp: 9.0 mmHg). Conclusion: PICU patients are a unique subset of the general population, and general age-normal vitals may not be suitable as target values indicating physiologic stability at discharge. Age-normal vitals that were specifically derived from the medical-to-physical discharge window of ICU patients may be more appropriate targets for 'acceptable' physiologic state for critical care patients. Going beyond simple age bins, an RNN model can provide more personalized target values.