PyData Berlin 2017
2017-06-23..2017-06-25
Berlin, Germany
[PyData Berlin 2014, PyData Berlin 2015, PyData Berlin 2016, PyData Berlin 2017, PyData Berlin 2018]
| title | author | length |
|---|---|---|
| Ethics in Machine Learning Panel | 56m | |
Panel records a practical debate about harms, fixes and governance with concrete examples and proposals: Tay chatbot turning racist, Google and Flickr… | ||
| Ethics in Machine Learning Panel Q&A | 33m | |
Panel records a wide-ranging debate that names concrete failures and proposals: cites Tay chatbot turning racist and Google and Flickr mislabeling… | ||
| Keynote - The Future of Cybersecurity Needs You, Here is Why. | Verónica Valeros | 55m |
Argues that two concrete gaps decide outcomes: slow real‑time response and slow post‑compromise discovery. Describes a pipeline using proxy logs and recently… | ||
| Keynote Q&A – Verónica Valeros | Verónica Valeros | 46m |
Argues that machine learning is essential to sift terabytes of network data and describes a layered ML pipeline of tiny, fast anomaly detectors that emit… | ||
| Keynote - Are many of your worries about AI wrong? | Toby Walsh | 59m |
Argues superintelligence is decades away—perhaps 50–100 years—and current danger comes from autonomy, not sentience. Gives concrete examples: a Tesla autopilot… | ||
| Keynote Q&A – Toby Walsh | Toby Walsh | 42m |
Covers concrete incidents and technical choices: the Tesla autopilot fatality discussion and legal accountability, the claim that humans cannot reliably retake… | ||
| Keynote - Natural Language Processing: Challenges and Next Frontiers | Barbara Plank | 55m |
Barbara Plank is tenured Assistant Professor in Natural Language Processing at the University of Groningen, The Netherlands.
Her research focuses on… | ||
| Keynote Q&A – Barbara Plank | Barbara Plank | 31m |
Argues that combining deep multi-task learning with oddly available data sources yields practical gains: word embeddings pre-initialization on huge unlabeled… | ||
| Data Science & Data Visualization in Python | Radovan Kavicky | 33m |
Argues that open data already exists in large unused volume and that turning it into policy requires concrete tools and deliverables: Anaconda and Jupyter… | ||
| Best Practices for Debugging | Kristian Rother | 01h 27m |
A hands‑on course that walks through a broken Python pipeline parsing UniProt‑style FASTA files and shows concrete fixes: a missing colon causing a syntax… | ||
| Engage the Hyper-Python | Daniele Rapati | 24m |
Shows concrete checks and outcomes: profile with cProfile and a system monitor like htop to decide whether CPU, memory, disk, or bandwidth is the bottleneck… | ||
| Where are we looking? Predicting human gaze using deep networks. | Oliver Eberle | 38m |
Claims a low‑level intensity contrast model matches or beats many classic saliency methods and that a learned Gabor bank narrows the gap to high‑level models… | ||
| Text similiarity with the next generation of word embeddings in Gensim | Lev Konstantinovskiy | 40m |
Shows concrete choices and outcomes: uses Gensim with Google's Word2vec, Facebook FastText, VarEmbed and WordRank; on a 120k‑word Pride and Prejudice corpus… | ||
| Analysing user comments with Doc2Vec and Machine Learning classification | Robert Meyer | 35m |
Doc2Vec training on scraped German comment threads yields concrete artifacts: Doc2Vec document embeddings, word vectors, and a linear classifier pipeline. Data… | ||
| Compositional distributional semantics for modelling natural language | Thomas Kober | 36m |
Summarises concrete composition methods and what actually works: compares pointwise addition and multiplication (surprisingly hard to beat), matrix/tensor… | ||
| Conversational AI: Building clever chatbots | Tom Bocklisch | 30m |
Open source Rasa is the concrete product: a pipeline using scikit‑learn, SpaCy and character n‑grams with a conditional random field for NER, SVM or simple… | ||
| Semi-Supervised Bootstrapping of Relationship Extractors | David Soares Batista | 39m |
Claims that replacing TF‑IDF with distributional word embeddings improves semi‑supervised bootstrapping for relationship extraction, extracting more correct… | ||
| Is That a Duplicate Quora Question? | Abhishek Thakur | 34m |
Uses the Quora dataset and walks through concrete features and models: TF‑IDF + SVD, fuzzy string scores, Word2Vec/GloVe sentence vectors, TF‑IDF SVD Word2Vec… | ||
| Developments in Test-Driven Data Analysis | Nick Radcliffe | 39m |
Test‑driven data analysis applies concrete checks: reference tests capture inputs code and expected outputs, while automatic constraint discovery generates… | ||
| Evaluating Topic Models | Matti Lyra | 45m |
Shows which evaluation steps actually change decisions: uses pyLDAvis and gensim for visualization, computes coherence scores and word‑intrusion tests, and… | ||
| A word is worth a thousand pictures: Convolutional methods for text | Tal Perry | 32m |
Claims that convolutional nets beat RNNs on speed and deployability backed by concrete implementations in TensorFlow including DenseNet adaptation… | ||
| Gold standard data: lessons from the trenches | Miroslav Batchkarov | 36m |
Three case studies show exactly what matters: concrete labels, tooling and measured outcomes. Word-embedding evaluation reveals huge human variance (ratings… | ||
| The path between developing and serving machine learning models. | Adrin Jalali | 58m |
Deploys a Kubernetes cluster (kops on AWS), installs PipelineIO components — Spark, HDFS, JupyterHub, a Java PMML server plus a Python prediction service — and… | ||
| Advanced Metaphors in Coding with Python | James Powell | 01h 25m |
Compressed, code-first walkthrough of how Python features actually get used: concrete demos of metaclass hooks including the build_class hook and how ABCMeta… | ||
| Leveling up your Jupyter notebook skills | Gerrit Gruben | 01h 44m |
Shows concrete, repeatable steps: use modal editor like VIM keybindings, create isolated conda envs and port‑forward a remote kernel, and install nbextensions… | ||
| Introduction to Julia for Scientific Computing and Data Science | David Higgins, Robert Schwarz | 01h 27m |
Claims that Julia lets one write high‑level code but run at compiled speed, demonstrated on Julia 0.6 inside Jupyter notebooks with precompile overhead noted… | ||
| Pandas from the Inside / "Big Pandas" | Stephen Simmons | 01h 17m |
Pandas internals get examined: DataFrames are blocks of homogeneous numpy arrays managed by a block manager, so pulling a Series is often microsecond-fast… | ||
| Introduction to Data-Analysis with Pandas | Alexander CS Hendorf | 01h 24m |
Covers a hands‑on Pandas workflow using Jupyter notebooks and examples that read CSV/Excel/JSON/SQL/HTML via read_csv / to_excel on a ~1,000‑row, 7‑column… | ||
| Topic Modelling (and more) with NLP framework Gensim | Bhargav Srinivasa Desikan | 48m |
Demo shows a complete pipeline: spaCy preprocessing (lemmatization, curated stopwords, bigrams) turns messy news text into a gensim-ready corpus; gensim runs… | ||
| Social Networks and Protest Participation: Evidence from 130 Million Twitter Users | Jonathan Ronen | 26m |
Claims that network position predicts protest attendance using 130 million Twitter users as the crawl and Python tooling plus the Twitter REST API to build a… | ||
| Data Science for Digital Humanities: Extracting meaning from Images and Text | Hendrik Heuer | 37m |
Claim: Python pipelines make large-scale image and text work practical. Concrete tools and results: scikit-learn, Keras, gensim, ImageNet 1.2M images… | ||
| AI assisted creativity | Roelof Pieters | 36m |
Description
A new wave of creative applications of AI has arrived, making science fiction authors struggle to keep up with reality. Recent advances in Deep… | ||
| Spying on my Network for a Day: Data Analysis for Networks | Aisha Bello | 30m |
Collected one day of home packet captures (about 354 total) producing over 450,000 packets and processed with Wireshark then exported to CSV for visualization… | ||
| Biases are bugs: algorithm fairness and machine learning ethics | Françoise Provencher | 38m |
Claim: biases are bugs that must be found, fixed, and learned from. Concrete methods include stress cases that enumerate worst harms, automated association… | ||
| Fairness and transparency in machine learning: Tools and techniques | Andreas Dewes | 42m |
Shows a concrete pipeline: use Python and scikit-learn to train logistic classifiers on New York stop-and-frisk records, measure disparate impact (e g 50% of… | ||
| Deep Learning for detection on a phone | Irina Vidal Migallón | 38m |
Pipeline shows how a detector gets from prototype to phone: configuration-driven deployment file, model definition converted to convolutional-only layers… | ||
| Towards Pythonic Innovation in Recommender Systems | Carlotta Schatten | 38m |
Claims that Python libraries (Surprise, LightFM, MLlib/ALS, FM‑light) are surveyed and found incomplete: LightFM supports context and OpenMP parallelization… | ||
| “Which car fits my life?” - mobile.de’s approach to recommendations | Florian Wilhelm, Arnab Dutta | 43m |
mobile.de runs a hybrid recommender that combines collaborative matrix factorization with content vectors to handle a rapidly changing 1.6 million-item… | ||
| On Bandits, Bayes, and swipes: gamification of search | Stefan Otte | 36m |
A swipe‑based car search app actively learns a user’s taste from a few gestures by treating features (brand, body, segment) as separate multi-armed bandits and… | ||
| Machine Learning to moderate ads | Vaibhav Singh, Jaroslaw Szymczak | 33m |
Claims a deployed system that moderates 100+ million classifieds per month using concrete choices: feature hashing (about 200,000 buckets for description and… | ||
| Large Scale Vandalism Detection in Knowledge Bases | Alexey Grigorev | 32m |
Wikidata revision dump contains 72 million edits totalling half a terabyte of revisions with only 25 basis points rollback rate so the task is extreme class… | ||
| Open Data Use Cases | Ulrike Thalheim | 45m |
Overview of current German open data landscape notes that the German Open Data Law passed 2017 regains attention and may increase accessible administrative… | ||
| Clean Code in Jupyter notebooks, using Python | Volodymyr (Vlad) Kazantsev | 39m |
Advocates concrete rules: keep each notebook small—four to ten cells per hypothesis—and make each cell produce one output per cell so readers follow the story… | ||
| Fast Multidimensional Signal Processing using Julia with Shearlab.jl | Hector Andrade Loarca | 35m |
Shearlab.jl implements the Shearlet Transform and wavelets in Julia and shows concrete code for filtering, FFT use, GPU parallelism and iterative hard… | ||
| Polynomial Chaos: A technique for modeling uncertainty | Emily Gorcenski | 38m |
Polynomial chaos represents any finite‑variance random parameter as a deterministic series of orthogonal polynomials, computes coefficients by Galerkin… | ||
| TNaaS - Tech Names as a Service | Vincent D Warmerdam | 40m |
A web service generates Pokemon‑like names and actually runs at https://dyccxmwpz9.execute-api.eu-west-1.amazonaws.com/pokemon/poke-names/v2 with the domain… | ||
| Finding Lane Lines for Self Driving Cars | Ross Kippenbrock | 38m |
A concrete pipeline uses Python and OpenCV to undistort, warp to bird’s-eye view, apply Gaussian blur, color thresholds, Canny and Sobel edge detection… | ||
| Building smart IoT applications with Python and Spark | Rafael Schultze-Kraft | 39m |
Description
In this talk I will present how we use Python, PySpark and AWS as our preferred data science stack for the Internet of Things, which allows us to… | ||
| Introduction to Search | Sirin Odrowski | 38m |
WordPress.com builds a reader search on Elasticsearch that indexes content from 90 million posts per month, serving about 60 million queries/day and ~12… | ||
| Patterns for Collaboration Between Data Scientists and Software Engineers | Karolina Alexiou | 37m |
Describes practical collaboration patterns between data scientists and software engineers using CI-driven model pipelines that run parameterized Jupyter… | ||
| Kickstarting projects with Cookiecutter | Raphael Pierzina | 28m |
Cookiecutter is a CLI that installs from PyPI, clones templates over git or mercurial, prompts for values and renders files with Jinja2; the project reports… | ||
| Patsy: The Lingua Franca to and from R | Max Humber | 30m |
Shows how Patsy + scikit-learn reproduces R formula convenience by building design matrices, handling categorical variables and fitting three flavors of… | ||
| Introduction to Machine Learning with H2O and Python | Jo-fai Chow | 01h 10m |
Shows runnable Jupyter notebooks that actually start a local H2O JVM cluster and connect to a cloud node, import CSVs or convert pandas frames, and run models… | ||
| Introductory tutorial on data exploration and statistical models | Alexandru Agachi | 01h 24m |
Hands-on pipeline that loads four heart-disease files, concatenates them, and cleans data with pandas/StatsModels/scikit-learn in an Anaconda/Jupyter… | ||
| Lightning Talks | 39m | |
Brief lightning sequence lists concrete claims and demos: a mammogram pipeline uses candidate extraction plus neural nets that now match radiologist… | ||
| Blockchains for Artificial Intelligence | Trent McConaghy | 49m |
Claims that blockchains solve AI’s data problems get concrete here: IPDB and BigchainDB provide immutable audit trails and provenance; prototypes include an AV… | ||
| nothing matches | ||