The Graybeards tell us that all of C++ rests on the back of templating, to which the skeptic responds, “Then what supports the templating?” For a true metaprogrammer like myself, the answer is obvious: “It’s templates all the way down.” Templates of templates compiling templates and begetting more templates, or at least that’s what I used to believe… In this post, we’ll build a simple x86 assemble
Password recovery Contents Password recovery Data transformation Saving password into the network Recovering password from the network Test it using Monte Carlo Possible problems Summary Download script In this article we are going to build a simple neural network that will recover password from a broken one. If you aren’t familiar with a Discrete Hopfield Network algorithm, you can read this arti
Basically, if you feed it an image, say the classic 3x512x512 Lenna image: And you run this code on it: require 'dpnn' require 'image' img = image.lena() loc = torch.Tensor{0,0} -- 0.0 is the center of the image sg = nn.SpatialGlimpse(64, 3, 2) output = sg:forward{img, loc} print(output:size()) -- 9 x 64 x 64 -- unroll the glimpse onto its different depths outputs = torch.chunk(output, 3) display
CNNは画像認識の分野で驚異的な精度を誇るディープラーニングのアルゴリズムのひとつであるものの、ぱっと見がとても複雑な構造をしているため、実装するのも大変そうです。 実際、ネットや文献上で見られる多くのCNNの実装は、Theano (pythonのライブラリ)の自動微分機能を使っていたり、MATLABの組み込み関数を使っているものがほとんどです。 そのためか、きちんと forward propagation & backpropagation を数式で書き下している文献はないように思いました。(もちろん、楽に実装できるならばそれはそれで素晴らしいことです。) そこで、どうすれば CNN を実装するための数式を書き下せるのか、レイヤーごとに分けて導出していきたいと思います。 まず、CNN がどんな層に分解できるのかについて。これは、下記の3つで表せるでしょう。 Convolution Lay
Netflix is pleased to announce the open source release of our x.509 certificate orchestration framework: Lemur! The Challenge of Certificate ManagementPublic Key Infrastructure is a set of hardware, software, people, policies, and procedures needed to create, manage, distribute, use, store, and revoke digital certificates and manage public-key encryption. PKI allows for secure communication by est
EngineeringCounting ObjectsThe Systems Team at GitHub works to solve complex bugs and performance bottlenecks at the lowest levels of our infrastructure. Over the past two years we've undertaken a major project… The Systems Team at GitHub works to solve complex bugs and performance bottlenecks at the lowest levels of our infrastructure. Over the past two years we’ve undertaken a major project to i
Morphological image processing Here's a cool little topic. Let's think about how to extract borders in an image. I'm sure you could think of hundreds of heuristics, but I'd like to discuss one particular heuristic that, indirectly, leads to some pretty cool results relating to cellular automata. The technique is called "morphology", and it's one of those ideas which starts out with very simple rul
Summary: Using the technique from the previous post, here are three space leaks I found. Every large Haskell program almost inevitably contains space leaks. This post examines three space leaks I found while experimenting with a space-leak detection algorithm. The first two space leaks have obvious causes, but I remain mystified by the third. Hoogle leak 1The motivation for looking at space leak d
Here we are luminance-0.1 was released yesterday night, along with luminance-samples-0.1! I’ll need to enhance the documentation and add directions so that people don’t feel too overwhelmed. I’m also going to write a wiki to help people get their mind wrapped around luminance. If you think something is missing; if you think something could be enhanced; or if you’ve found a bug, please, feel free t
The Haskell programming language community. Daily news and info about all things Haskell related: practical stuff, theory, types, libraries, jobs, patches, releases, events and conferences and more...
The latest news from Google on open source releases, major projects, events, and student outreach programs. At Google, we think that internet users’ time is valuable, and that they shouldn’t have to wait long for a web page to load. Because fast is better than slow, two years ago we published the Zopfli compression algorithm. This received such positive feedback in the industry that it has been in
このところ、たびたび NumPy 後継が...とか 並列処理が...という話を聞くので、この秋 注目の多次元配列パッケージをまとめたい。 バックエンド系 NumPy のように数値計算処理を自前で実装しているパッケージ。 DyND Blaze プロジェクトのひとつ。C++ 実装 + Python バインディング。GitHub にいくつか Example があがっているが、複合型やカテゴリカル型、GroupBy 操作がサポートされていて熱い。ラベルデータも NumPy より簡単に実装できそうだ。 speakerdeck.com 並列分散系 自身では直接 数値計算処理を行わず、バックエンド ( 主に NumPy )を利用して並列/分散処理を行うパッケージ。1 物理PC/複数コアでの並列計算を主用途とし、NumPy, pandas では少し苦しいが PySpark などを使うほどじゃない...とい
リリース、障害情報などのサービスのお知らせ
最新の人気エントリーの配信
j次のブックマーク
k前のブックマーク
lあとで読む
eコメント一覧を開く
oページを開く