- Deep Learning for Analytical Engine — This repository contains an implementation of a convolutional neural network as a program for Charles Babbage’s Analytical Engine, capable of recognizing handwritten digits to a high degree of accuracy (98.5% if provided with a sufficient amount of training data and left running sufficiently long).
- Supervisor Trees in Go — A well-structured Erlang program is broken into multiple independent pieces that communicate via messages, and when a piece crashes, the supervisor of that piece automatically restarts it. […] Even as I have been writing suture, I have on occasion been astonished to flip my screen over to the console of Go program I’ve written with suture, and been surprised to discover that it’s actually been merrily crashing away during my manual testing, but soldiering on so well I didn’t even know.
- How to Avoid Brittle Code — If it hurts, do it more often.
- Developing Quantum Annealer Driven Data Discovery (Joseph Dulny III, Michael Kim) — In this paper, we gain novel insights into the application of quantum annealing (QA) to machine learning (ML) through experiments in natural language processing (NLP), seizure prediction, and linear separability testing.
"software" entries
Four short links: 30 March 2016
Deep Babbage, Supervisors in Go, Brittle Code, and Quantum NLP
Four short links: 24 March 2016
Work and Home Github, Museum Data, Bandwidth Incentives, and Motion Design
- Maintain Separate Github Accounts — simple advice.
- Cooper-Hewitt Pen Data — anonymized data from the Cooper-Hewitt design museum’s fantastic pen.
- Zero Rating’s Problem — Wikipedia was zero-rated for Angola, so Angolans began swapping movies via Wikipedia. Zero rating (“no data charge for this service”) is an incentive to use the site, not necessarily for the purpose intended.
- Motion Design is the Future of UI — Motion tells stories. Everything in an app is a sequence, and motion is your guide. Someone caught the animations and transitions bug.
Four short links: 3 March 2016
Tagging People, Maintenance Anti-Pattern, Insourced Brains, and Chat UI
- Human Traffickers Using RFID Chips (NPR) — It turns out this 20-something woman was being pimped out by her boyfriend, forced to sell herself for sex and hand him the money. “It was a small glass capsule with a little almost like a circuit board inside of it,” he said. “It’s an RFID chip. It’s used to tag cats and dogs. And someone had tagged her like an animal, like she was somebody’s pet that they owned.”
- Software Maintenance is an Anti-Pattern — Governments often use two anti-patterns when sustaining software: equating the “first release” with “complete” and moving to reduce sustaining staff too early; and how a reduction of staff is managed when a reduction in budget is appropriate.
- Cloud Latency and Autonomous Robots (Ars Technica) — “Accessing a cloud computer takes too long. The half-second time delay is too noticeable to a human,” says Ishiguro, an award-winning roboticist at Osaka University in Japan. “In real life, you never wait half a second for someone to respond. People answer much quicker than that.” Tech moves in cycles, from distributed to centralized and back again. As with mobile phones, the question becomes, “what is the right location for this functionality?” It’s folly to imagine everything belongs in the same place.
- Chat as UI (Alistair Croll) — The surface area of the interface is almost untestable. The UI is the log file. Every user interaction is also a survey. Chat is a great interface for the Internet of Things. It remains to be seen how many deep and meaningfuls I want to have with my fridge.
Four short links: 23 February 2016
AI or IA, Retro Chatbots, Science by Software, and Spec as Test Oracle
- Doing Something For Me vs Allowing Me To Do Even More (Matt Webb) — nails the split in startups. Come on, valley kids … do you want diapers or do you want superpowers?
- Paul Ford on Racter — But don’t get too ahead of things. Using Racter is not as different from using Siri as you might expect. It’s just that Siri has petabytes of stuff in her brain, whereas Racter has a floppy’s worth. Computers have changed a ton in the last 30 years, humans barely at all. Don’t mistake their progress for ours. We’ve learned how to talk to computers, and they’ve learned how to pretend to understand us. Useful when driving. People love chatting with their Amazon Echo. But the conversation still doesn’t really mean anything.
- Accelerating Science: A Computing Research Agenda (PDF) — Siri thinks I want to tell telemarketers to “duck off,” while researchers look to automated hypothesis generation, experiment design, results analysis, and knowledge integration.
- Not Quite So Broken TLS (Adrian Colyer) — instead of ad-hoc codery, A precise and testable specification (in this case for TLS) that unambiguously determines the set of behaviours it allows (and hence also what it does not). The specification should also be executable as a test oracle, to determine whether or not a given implementation is compliant. The paper outlines this for TLS, but I see formal methods growing in importance in coming years. We can’t build an airport with cardboard on a swamp. In this metaphor, cardboard represents our ad hoc dev practices and the swamp is our platform of crap code. The airport is … look, never mind, I’ll work on the metaphor. Read the paper.
Four short links: 29 January 2016
LTE Security, Startup Tools, Security Tips, and Data Fiction
- LTE Weaknesses (PDF) — ShmooCon talk about how weak LTE is: a lot of unencrypted exchanges between handset and basestation, cheap and easy to fake up a basestation.
- Analyzo — Find and Compare the Best Tools for your Startup it claims. We’re in an age of software surplus: more projects, startups, apps, and tools than we can keep in our heads. There’s a place for curated lists, which is why every week brings a new one.
- How to Keep the NSA Out — NSA’s head of Tailored Access Operations (aka attacking other countries) gives some generic security advice, and some interesting glimpses. “Don’t assume a crack is too small to be noticed, or too small to be exploited,” he said. If you do a penetration test of your network and 97 things pass the test but three esoteric things fail, don’t think they don’t matter. Those are the ones the NSA, and other nation-state attackers will seize on, he explained. “We need that first crack, that first seam. And we’re going to look and look and look for that esoteric kind of edge case to break open and crack in.”
- The End of Big Data — future fiction by James Bridle.
Matthew Berggren on making electronics accessible
The O’Reilly Hardware Podcast: Better ways to design electronics.
Subscribe to the O’Reilly Hardware Podcast for insight and analysis about the Internet of Things and the worlds of hardware, software, and manufacturing.
In our new episode of the Hardware Podcast, David Cranor and I talk with Matthew Berggren, who at the time the interview was conducted last December was senior director of product at Supplyframe. (Berggren is now director of Autodesk Circuits at Autodesk.)
Our discussion focuses on the need for abstracted modules and better metadata in electronics. Berggren gets to the root of it here:
There are 30 software developers for every hardware engineer in the world. That’s not only a tremendous bottleneck, but if you accept the premise that the next generation of products are going to be some hybrid of hardware and software—and really, hardware is the means to interact with the real world, and I want to write software applications that will interact with the real world—then there is this massive blue ocean out there that should present tremendous opportunity to semiconductor manufacturers, or anyone else who wants to get into that space.
Four short links: 18 January 2016
Machine Learning Technical Debt, Audio Matching, Self-Tracking Research, and Baidu's Open Source Deep Learning Code
- Hidden Technical Debt in Machine Learning Systems (PDF) — We explore several ML-specific risk factors to account for in system design. These include boundary erosion, entanglement, hidden feedback loops, undeclared consumers, data dependencies, configuration issues, changes in the external world, and a variety of system-level anti-patterns.
- Large-Scale Content-Based Matching of Midi and Audio Files — We present a system that can efficiently match and align MIDI files to entries in a large corpus of audio content based solely on content, i.e., without using any metadata.
- Critical Social Research on Self-Tracking — I am currently working on an article that is a comprehensive review of both literatures, in the attempt to outline what each can contribute to understanding self-tracking as an ethos and a practice, and its wider sociocultural implications. Here is a reading list of the work from critical social researchers that I am aware of. Trigger warning: phrases like “The discursive construction of student subjectivities.”
- Warp-CTC — Baidu’s open source deep learning code. Connectionist Temporal Classification is a loss function useful for performing supervised learning on sequence data, without needing an alignment between input data and labels.
Charles Fracchia on a new breed of biologists
The O’Reilly Hardware Podcast: The merging worlds of software, hardware, and biology.
Subscribe to the O’Reilly Hardware Podcast for insight and analysis about the Internet of Things and the worlds of hardware, software, and manufacturing.
In this new episode of the Hardware Podcast—which features our first discussion focusing specifically on synthetic biology—David Cranor and I talk with Charles Fracchia, an IBM Fellow at the MIT Media Lab and founder of the synthetic biology company BioBright.
Discussion points:
- The blurring of the lines between biology, software development, hardware engineering, and electrical engineering
- BioBright’s efforts to create hardware and software tools to reinvent the way biology is done in a lab
- The most prominent market forces in biology today (especially healthcare)
- How experiments conducted using Arduino or Raspberry Pi devices are impacting synthetic biology
- Pembient’s synthetic rhino horns
Four short links: 28 July 2015
Auto-Remediation, Fast and Good, Life's Game of Conway, and Self-Assembly Lab
- Nurse at LinkedIn — automating the responses to alerts.
- Moving Fast With High Code Quality (Quora) — Lots of practical detail about how they combine speed with quality.
- John Horton Conway (The Guardian) — These were two separate areas of study that Conway had arrived at by two different paths. So, there’s no reason for them to be linked. But somehow, through the force of his personality, and the intensity of his passion, he bent the mathematical universe to his will. Fascinating profile, taken from a new book.
- MIT Self-Assembly Lab — multi-material 3D/4D printing, advances in materials science, and new capabilities in simulation/optimization software […] made it possible to fully program a wide range of materials to change shape, appearance, or other property, on demand.
Four short links: 10 July 2015
King Rat Brain, Emojactions, Dead Eye, and Cloud Value
- Computer of Wired-Together Rat Brains — this is ALL THE AMAZING. a Brainet that allows three monkeys connected at the brain to control a virtual arm on screen across three axes. […] Nicolelis said that, essentially, he created a “classic artificial neural network using brains.” In that sense, it’s not artificial at all. (via Slashdot)
- Reactions — Slack turns emoji into first-class interactions. Genius!
- Pixar’s Scientific Method — If you turn your head without moving your eyes first, it looks like you’re dead. Now there’s your uncanny valley.
- AWS CodePipeline — latest in Amazon’s build-out of cloud tools. Interchangeable commodity platforms regaining lockin via higher-order less-interchangeable tooling for deployment, config, monitoring, etc.