Towards a method for fixing machine learning’s persistent and catastrophic blind spots

This entry was posted in Interesting Phenomena on by .

An adversarial preturbation is a small, human-imperceptible change to a piece of data that flummoxes an otherwise well-behaved machine learning classifier: for example, there’s a really accurate ML model that guesses which full-sized image corresponds to a small thumbnail, but if you change just one pixel in the thumbnail, the classifier stops working almost entirely.

In the midst of social media encouraging us to increasingly overshare our lives, a curious thing has happened: Journaling is back. And while the practice of jotting down your thoughts and plans in a private, analog medium is therapeutic, it can also be pretty productive. We’ve tracked down a few decidedly modern notebooks that have [‘]

Publisher: Boing Boing
Twitter: @BoingBoing
Reference: Visit Source

Other things to check out:

On the Rookout for live data: Instant observability to fix software bugs and open AI black boxes

Software bugs are a pain: buggy software can drop anything from your sales to aircraft in mid-flight. Debugging software is hard, tedious, and costs a fortune. A multitude of frameworks and processes have been created to facilitate software testing and ensure fewer bugs make it to production, and invariably they all fail from time to time. When this happens, the pain for software developers and the nail-biting for businesses starts. Developers have to find the source code that caused the bug, and execute this in a test environment that resembles the production one as closely as possible. The situation that caused the bug has to be recreated, too. The way this usually works is by adding logging statements and breakpoints in the code, and retracing execution in the code and its dependencies until the bug is located and can be fixed. Then the new code has to be rebuilt and redeployed in production. Frankly, it’s a pain just thinking about it, let alone having to go through this. Or Weis and Liran Haimovitch are two software engineers who have gone through this time after time, and felt the pain, so they decided to do something about it.

Publisher: ZDNet
Author: George Anadiotis
Twitter: @ZDNet
Reference: Visit Source

The Comedian Is in the Machine. AI Is Now Learning Puns

A pun generator might not sound like serious work for an artificial intelligence researcher’more the sort of thing knocked out over the weekend to delight the labmates come Monday. But for He He, who designed just that during her postdoc at Stanford, it’s an entry point to a devilish problem in machine learning. He’s aim is to build AI that’s natural and fun to talk to’bots that don’t just read us the news or tell us the weather, but can crack jokes or compose a poem, even tell a compelling story. But getting there, she says, runs up against the limits of how AI typically learns.

Publisher: WIRED
Date: 2019-05-03T11:00:00.000Z
Author: Gregory Barber
Twitter: @wired
Reference: Visit Source

BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules

The dichotomy of sweet and bitter tastes is a salient evolutionary feature of human gustatory system with an innate attraction to sweet taste and aversion to bitterness. A better understanding of molecular correlates of bitter-sweet taste gradient is crucial for identification of natural as well as synthetic compounds of desirable taste on this axis. While previous studies have advanced our understanding of the molecular basis of bitter-sweet taste and contributed models for their identification, there is ample scope to enhance these models by meticulous compilation of bitter-sweet molecules and utilization of a wide spectrum of molecular descriptors. Towards these goals, our study provides a structured compilation of bitter, sweet and tasteless molecules and state-of-the-art machine learning models for bitter-sweet taste prediction (BitterSweet). We compare different sets of molecular descriptors for their predictive performance and further identify important features as well as feature blocks. The utility of BitterSweet models is demonstrated by taste prediction on large specialized chemical sets such as FlavorDB, FooDB, SuperSweet, Super Natural II, DSSTox, and DrugBank. To facilitate future research in this direction, we make all datasets and BitterSweet models publicly available, and present an end-to-end software for bitter-sweet taste prediction based on freely available chemical descriptors.

Publisher: Scientific Reports
Date: 2019-05-09
Author: 2019 The Author s
Twitter: @SciReports
Reference: Visit Source

Greetings Earthlings: All systems on halt. The data presented above may one day be zapped to another dimension. Just thought you should be aware. Dude, there was a blue light over there just now.