"machine learning" entries

Risto Miikkulainen on evolutionary computation and making robots think for themselves

The O'Reilly Radar Podcast: Evolutionary computation, its applications in deep learning, and how it's inspired by biology.

Subscribe to the O’Reilly Radar Podcast to track the technologies and people that will shape our world in the years to come: Stitcher, TuneIn, iTunes, SoundCloud, RSS

Haeckel_Orchidae

In this week’s episode, David Beyer, principal at Amplify Partners, co-founder of Chart.io, and part of the founding team at Patients Know Best, chats with Risto Miikkulainen, professor of computer science and neuroscience at the University of Texas at Austin. They chat about evolutionary computation, its applications in deep learning, and how it’s inspired by biology.

Finding optimal solutions

We talk about evolutionary computation as a way of solving problems, discovering solutions that are optimal or as good as possible. In these complex domains like, maybe, simulated multi-legged robots that are walking in challenging conditions—a slippery slope or a field with obstacles—there are probably many different solutions that will work. If you run the evolution multiple times, you probably will discover some different solutions. There are many paths of constructing that same solution. You have a population and you have some solution components discovered here and there, so there are many different ways for evolution to run and discover roughly the same kind of a walk, where you may be using three legs to move forward and one to push you up the slope if it’s a slippery slope.

You do (relatively) reliably discover the same solutions, but also, if you run it multiple times, you will discover others. This is also a new direction or recent direction in evolutionary computation—that the standard formulation is that you are running a single run of evolution and you try to, in the end, get the optimum. Everything in the population supports finding that optimum.

Read more…

Comment
Four short links: 11 February 2016

Four short links: 11 February 2016

Surviving Crashes, Thumbs-Up Thumbs-Down Learning, Faster Homomorphic Encryption, and Nerdy V-Day Cards

  1. All File Systems are Not Created Equal: On the Complexity of Crafting Crash Consistent Applications (Paper a Day) — an important subject for me. BOB, the Block Order Breaker, is used to find out what behaviours are exhibited by a number of modern file systems that are relevant to building crash consistent applications. ALICE, the Application Level Intelligent Crash Explorer, is then used to explore the crash recovery behaviour of a number of applications on top of these file systems.
  2. BinaryNet: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1 (Arxiv) — instead of complex positive/negative floating-point weights, this uses +1 and -1 (which I can’t help but think of as “thumbs up”, “thumbs down”) to get nearly state-of-the-art results because a run-time, BinaryNet drastically reduces memory usage and replaces most multiplications by 1-bit exclusive-not-or (XNOR) operations, which might have a big impact on both general-purpose and dedicated Deep Learning hardware. GPLv2 code available.
  3. Microsoft Speeds Up Homomorphic Encryption (The Register) — homomorphic encryption lets databases crunch data without needing keys to decode it.
  4. Nerdy Valentine Cards (Evil Mad Scientist) — for a nerd in your life. (via Cory Doctorow)
Comment
Four short links: 5 February 2016

Four short links: 5 February 2016

Signed Filesystem, Smart Mirror, Deep Learning Tuts, and CLI: Miami

  1. Introducing the Keybase Filesystem — love that crypto is making its way into the filesystem.
  2. DIY Smart Bathroom Mirror — finally, someone is building this science-fiction future! (via BoingBoing)
  3. tensorflow tutorials — for budding deep learners.
  4. clmystery — a command-line murder mystery.
Comment
Four short links: 4 February 2016

Four short links: 4 February 2016

Shmoocon Video, Smart Watchstrap, Generalizing Learning, and Dataflow vs Spark

  1. Shmoocon 2016 Videos (Internet Archive) — videos of the talks from an astonishingly good security conference.
  2. TipTalk — Samsung watchstrap that is the smart device … put your finger in your ear to hear the call. You had me at put my finger in my ear. (via WaPo)
  3. Ecorithms — Leslie Valiant at Harvard broadened the concept of an algorithm into an “ecorithm,” which is a learning algorithm that “runs” on any system capable of interacting with its physical environment. Algorithms apply to computational systems, but ecorithms can apply to biological organisms or entire species. The concept draws a computational equivalence between the way that individuals learn and the way that entire ecosystems evolve. In both cases, ecorithms describe adaptive behavior in a mechanistic way.
  4. Dataflow/Beam vs Spark (Google Cloud) — To highlight the distinguishing features of the Dataflow model, we’ll be comparing code side-by-side with Spark code snippets. Spark has had a huge and positive impact on the industry thanks to doing a number of things much better than other systems had done before. But Dataflow holds distinct advantages in programming model flexibility, power, and expressiveness, particularly in the out-of-order processing and real-time session management arenas.
Comment
Four short links: 3 February 2016

Four short links: 3 February 2016

Security Forecast, Machine Learning for Defence, Retro PC Fonts, and Cognitive Psych Research

  1. Software Security Ideas Ahead of Their Time — astonishing email exchange from 1995 presaged a hell of a lot of security work.
  2. Doxxing Sherlock — Cory Doctorow’s ruminations on surveillance, Sherlock, and what he found in the Snowden papers. What he found included an outline of intelligence use of machine learning.
  3. Old-School PC Fonts — definitive collection of ripped-from-the-BIOS fonts from the various types of PCs. Your eyes will ache with nostalgia. (Or, if you’re a young gun, wondering how anybody wrote code with fonts like that) (my terminal font is VT220 because it makes me happy and productive)
  4. Cognitive Load: Brain GemsWe distill the latest behavioural economics & consumer psychology research down into helpful little brain gems.
Comment
Four short links: 2 February 2016

Four short links: 2 February 2016

Fourth Industrial Revolution, Agent System, Evidence-Based Programming, and Deep Learning Service

  1. This is Not the Fourth Industrial Revolution (Slate) — the phrase “the fourth Industrial Revolution” has been around for more than 75 years. It first came into popular use in 1940.
  2. Huginn — MIT-licensed system for building agents that perform automated tasks for you online. They can read the Web, watch for events, and take actions on your behalf. Huginn’s Agents create and consume events, propagating them along a directed graph. Think of it as a hackable Yahoo! Pipes plus IFTTT on your own server.
  3. Evidence-Oriented Programming — design programming language syntax and features based on what research shows works. They tested Perl and Java, found apparently not detectably easier to use for novices than a language that my student at the time, Susanna Kiwala (formerly Siebert), created by essentially rolling dice and picking (ridiculous) symbols at random.
  4. Deep Detect — open source deep learning service.
Comment
Four short links: 28 January 2016

Four short links: 28 January 2016

Augmented Intelligence, Social Network Limits, Microsoft Research, and Google's Go

  1. Chimera (Paper a Day) — the authors summarise six main lessons learned while building Chimera: (1) Things break down at large scale; (2) Both learning and hand-crafted rules are critical; (3) Crowdsourcing is critical, but must be closely monitored; (4) Crowdsourcing must be coupled with in-house analysts and developers; (5) Outsourcing does not work at a very large scale; (6) Hybrid human-machine systems are here to stay.
  2. Do Online Social Media Remove Constraints That Limit the Size of Offline Social Networks? (Royal Society) — paper by Robin Dunbar. Answer: The data show that the size and range of online egocentric social networks, indexed as the number of Facebook friends, is similar to that of offline face-to-face networks.
  3. Microsoft Embedding ResearchTo break down the walls between its research group and the rest of the company, Microsoft reassigned about half of its more than 1,000 research staff in September 2014 to a new group called MSR NExT. Its focus is on projects with greater impact to the company rather than pure research. Meanwhile, the other half of Microsoft Research is getting pushed to find more significant ways it can contribute to the company’s products. The challenge is how to avoid short-term thinking from your research team. For instance, Facebook assigns some staff to focus on long-term research, and Google’s DeepMind group in London conducts pure AI research without immediate commercial considerations.
  4. Google’s Go-Playing AIThe key to AlphaGo is reducing the enormous search space to something more manageable. To do this, it combines a state-of-the-art tree search with two deep neural networks, each of which contains many layers with millions of neuron-like connections. One neural network, the “policy network,” predicts the next move, and is used to narrow the search to consider only the moves most likely to lead to a win. The other neural network, the “value network,” is then used to reduce the depth of the search tree — estimating the winner in each position in place of searching all the way to the end of the game.
Comment
Four short links: 20 January 2016

Four short links: 20 January 2016

Rules-Based Distributed Code, Open Source Face Recognition, Simulation w/Emoji, and Berkeley's AI Materials

  1. Experience with Rules-Based Programming for Distributed Concurrent Fault-Tolerant Code (A Paper a Day) — To demonstrate applicability outside of the RAMCloud system, the team also re-wrote the Hadoop Map-Reduce job scheduler (which uses a traditional event-based state machine approach) using rules. The original code has three state machines containing 34 states with 163 different transitions, about 2,250 lines of code in total. The rules-based re-implementation required 19 rules in 3 tasks with a total of 117 lines of code and comments. Rules-based systems are powerful and underused.
  2. OpenFace — open source face recognition software using deep neural networks.
  3. Simulating the World in Emoji — fun simulation environment in the browser.
  4. Berkeley’s Intro-to-AI MaterialsWe designed these projects with three goals in mind. The projects allow students to visualize the results of the techniques they implement. They also contain code examples and clear directions, but do not force students to wade through undue amounts of scaffolding. Finally, Pac-Man provides a challenging problem environment that demands creative solutions; real-world AI problems are challenging, and Pac-Man is, too.
Comment
Four short links: 18 January 2016

Four short links: 18 January 2016

Machine Learning Technical Debt, Audio Matching, Self-Tracking Research, and Baidu's Open Source Deep Learning Code

  1. 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.
  2. Large-Scale Content-Based Matching of Midi and Audio FilesWe 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.
  3. Critical Social Research on Self-TrackingI 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.”
  4. 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.
Comment
Four short links: 5 January 2016

Four short links: 5 January 2016

Inference with Privacy, RethinkDB Reliability, T-Mobile Choking Video, and Real-Time Streams

  1. Privacy-Preserving Inference of Social Relationships from Location Data (PDF) — utilizes an untrusted server and computes the building blocks to support various social relationship studies, without disclosing location information to the server and other untrusted parties. (via CCC Blog)
  2. Jepson takes on Rethink — the glowingest review I’ve seen from Aphyr. As far as I can ascertain, RethinkDB’s safety claims are accurate.
  3. T-Mobile’s BingeOn `Optimization’ Is Just Throttling (EFF) — T-Mobile has claimed that this practice isn’t really “throttling,” but we disagree. It’s clearly not “optimization,” since T-Mobile doesn’t alter the actual content of the video streams in any way.
  4. qminer — BSD-licensed data analytics platform for processing large-scale, real-time streams containing structured and unstructured data.
Comment