platform/research augmentation, predictions, inferences
Latest releases by the giants
https://eng.uber.com/plato-research-dialogue-system/
The team @ Uber AI has developed the Plato Research Dialogue System, a platform for building, training, and deploying conversational AI agents that allows them to conduct state of the art research in conversational AI and quickly create prototypes and demonstration systems, as well as facilitate conversational data collection. They have designed Plato for both users with a limited background in conversational AI and seasoned researchers in the field by providing a clean and understandable design, integrating with existing deep learning and Bayesian optimization frameworks (for tuning the models), and reducing the need to write code.
https://venturebeat.com/2019/07/02/facebook-open-sources-dlrm-a-deep-learning-recommendation-model/
Last month, Facebook announced the open source release of Deep Learning Recommendation Model (DLRM), a state-of-the-art AI model for serving up personalized results in production environments. DLRM can be found on GitHub, and implementations of the model are available for Facebook’s PyTorch, Facebook’s distributed learning framework Caffe2, and Glow C++.
Right before Facebook’s introduction of DLRM, Amazon also announced the general availability of Amazon Personalize, an AWS service that facilitates the development of websites, mobile apps, and content management and email marketing systems that suggest products, provide tailored search results, and customize funnels on the fly.
https://eng.uber.com/uber-poet/
Before Swift 4.0, Uber determined that building many small Swift modules (~300) built with whole module optimization (WMO) mode was the fastest overall build mode for Uber. Swift 4.0 introduced a new batch build mode, which while advertised to work faster in most cases, was about 25 percent slower to build with our ~300 module configuration.
In order to test if refactoring the application part of their code into a few large modules would make the overall build time faster with the new batch mode but without actually refactoring it, Uber created Uber Poet, a mock application generator to simulate our target dependency structures. To enable others to benefit from our mock application generator, they have decided to open source it.
https://newsroom.fb.com/news/2019/08/open-source-photo-video-matching/
A fortnight ago, FB open-sourced two technologies that detect identical and nearly identical photos and videos — sharing some of the tech they use to fight abuse on their platform with others who are working to keep the internet safe. These algorithms will be open-sourced on GitHub so that their industry partners, smaller developers and non-profits can use them to more easily identify abusive content and share hashes — or digital fingerprints — of different types of harmful content. For those who already use their own or other content matching technology, these technologies are another layer of defense and allow hash-sharing systems to talk to each other, making the systems that much more powerful
https://techcrunch.com/2019/07/19/googles-smily-is-reverse-image-search-for-cancer-diagnosis/
Spotting and diagnosing cancer is a complex and difficult process even for the dedicated medical professionals who do it for a living. A new tool from Google researchers could improve the process by providing what amounts to reverse image search for suspicious or known cancerous cells. But it’s more than a simple matching algorithm.
https://towardsdatascience.com/python-libraries-for-interpretable-machine-learning-c476a08ed2c7
As concerns regarding bias in artificial intelligence become more prominent it is becoming more and more important for businesses to be able to explain both the predictions their models are producing and how the models themselves work. Fortunately, there is an increasing number of python libraries being developed that attempt to solve this problem. In the above post, Rebecca Vickery has provided a brief guide to four of the most established packages for interpreting and explaining machine learning models.
https://techxplore.com/news/2019-07-technique-machine-behavior-human-infants.html
Despite the significant recent advances in the field of artificial intelligence (AI), most virtual agents still require hundreds of hours of training to achieve human-level performance in several tasks, while humans can learn how to complete these tasks in a few hours or less. Recent studies have highlighted two key contributors to humans' ability to acquire knowledge so quickly—namely, intuitive physics and intuitive psychology.
These intuition models, which have been observed in humans from early stages of development, might be the core facilitators of future learning. Based on this idea, researchers at the Korea Advanced Institute of Science and Technology (KAIST) have recently developed an intrinsic reward normalization method that allows AI agents to select actions that most improve their intuition models. In their paper, pre-published on arXiv, the researchers specifically proposed a graphical physics network integrated with deep reinforcement learning inspired by the learning behavior observed in human infants.
Historically, detection evasion has followed a common pattern: attackers would build new versions of their malware and test them offline against antivirus solutions. They’d keep making adjustments until the malware can evade antivirus products. Attackers then carry out their campaign knowing that the malware won’t initially be blocked by AV solutions, which are then forced to catch up by adding detections for the malware. In the cybercriminal underground, antivirus evasion services are available to make this process easier for attackers.
Microsoft Defender ATP’s Antivirus has significantly advanced in becoming resistant to attacker tactics like this. A sizeable portion of the protection they deliver are powered by machine learning models hosted in the cloud. The cloud protection service breaks attackers’ ability to test and adapt to their defenses in an offline environment, because attackers must either forgo testing, or test against our defenses in the cloud, where they can observe them and react even before they begin.
Hardening their defenses against adversarial attacks doesn’t end there. In the above blog they discuss a new class of cloud-based ML models that further harden their protections against detection evasion.
Late month, fans and haters of Elon Musk turned out by the thousands to watch an internet live-stream of the first public presentation by Neuralink, a company the Tesla billionaire formed two years ago with the dramatic (if not entirely new) goal of connecting people’s brains to computers.
The three-hour event was part marketing spectacle and part dry technical explainer. Musk and his team members described the brain-machine interface design they’re betting on, which will employ dozens of thin wires to collect signals in the brain, and which they want to try out on paralyzed people soon, so they can type with their minds. Their eventual aim is to connect those wires to a thought transmitter which tucks behind your ear like a hearing aid.
https://www.infoq.com/news/2019/07/google-tensorflow-text/
Google has released TensorFlow.Text (TF.Text), a new text-processing library for their TensorFlow deep-learning platform. The library allows several common text pre-processing activities, such as tokenization, to be handled by the TensorFlow graph computation system, improving consistency and portability of deep-learning models for natural-language processing (NLP).
https://eng.uber.com/uscs-apache-spark/
Apache Spark is a foundational piece of Uber’s Big Data infrastructure that powers many critical aspects of our business. Uber currently runs more than one hundred thousand Spark applications per day, across multiple different compute environments. Spark’s versatility, which allows them to build applications and run them everywhere that they need, makes this scale possible.
However, their ever-growing infrastructure means that their operational environments are constantly changing, making it increasingly difficult for both new and existing users to give their applications reliable access to data sources, compute resources, and supporting tools. Also, as the number of users grow, it becomes more challenging for the data team to communicate these environmental changes to users, and for them to understand exactly how Spark is being used.
Uber built the Uber Spark Compute Service (uSCS) to help manage the complexities of running Spark at this scale. This Spark-as-a-service solution leverages Apache Livy, currently undergoing Incubation at the Apache Software Foundation, to provide applications with necessary configurations, then schedule them across their Spark infrastructure using a rules-based approach.
A new experiment from MIT and Brown University researchers have added a capability to their ‘Northstar’ interactive data system that can “instantly generate machine-learning models” to use with their exiting data sets in order to generate useful predictions.
One example the researchers provide is that doctors could make use of the system to make predictions about the likelihood their patients have of contracting specific diseases based on their medial history. Or, they suggest, a business owner could use their historical sales data to develop more accurate forecasts, quickly and without a ton of manual analytics work.
https://hdsr.mitpress.mit.edu/pub/a7gxkn0a
The strategic role of data science teams in industry is fundamentally to help businesses to make smarter decisions. This includes decisions on minuscule scales, such as what fraction of a cent to bid on an ad placement displayed in a web browser, whose importance is only manifest when scaled by orders of magnitude through machine automation. But it also extends to singular, monumental decisions made by businesses, such as how to position a new entrant within a competitive market. In both regimes, the potential impact of data science is only realized when both humans and machine actors are learning from data and when data scientists communicate effectively to decision makers throughout the business.
A last month’s article from MIT press examines this dynamic through the instructive lens of the duality between inference and prediction. The article defines these concepts, which have varied use across many fields, in practical terms for the industrial data scientist. Through a series of descriptions, illustrations, contrasting concepts, and examples from the entertainment industry (box office prediction and advertising attribution), the author offers perspectives on how the concepts of inference and prediction manifest in the business setting