Züri ML #19: Word Embeddings, and Distributed ML
Details
ATTENTION: fully booked!
We're looking forward to this joint event with IBM Research Lab Zurich (http://www.research.ibm.com/labs/zurich/), and the Zürich Apache Spark Meetup (https://www.meetup.com/spark-zurich/) group
Preliminary program:
• Introduction to Word-Embeddings
Florian Schmidt (http://www.da.inf.ethz.ch/people/FlorianSchmidt/), ETH Zürich
Word embeddings map words to vectors in Eucledian space, thereby opening the door to a wide range of linear algebra techniques. Since the late 60's it is well known that unsupervised techniques can construct embeddings that preserve useful semantic properties such as pair-wise word similarity.
The recent surge of work in the field introduced both, a new regime for learning embeddings as well as new semantic properties that can be preserved under the embedding.
This talk will explain new prediction-based embedding techniques such as skip-gram (Mikolov et al.) , the word analogy task and why the methods' success on these tasks is actually a surprise – and maybe not even a novelty. Finally, some very recent applications of word embeddings will be presented.
• Leveraging Word-Embeddings for Natural and Semi-Natural Text Analytics
Dorothea Wiesmann (https://www.linkedin.com/pub/dorothea-wiesmann/2/b53/61), IBM Research
• Distributed Machine Learning with Apache Spark
Christina Heinze (https://www.linkedin.com/pub/christina-heinze/a3/9b4/a71), ETH Zürich
In the last few years there has been great interest in solving large-scale optimization and estimation problems. Parallelization has naturally emerged to leverage now-commonplace multi-core architectures to enable moderate sized problems to be solved quickly. Some datasets are large enough such that they are impractical to store and process on a single machine and so the problem must be solved in a distributed manner on a computing cluster. Two obvious questions arise: (1) how should the data and processing tasks be distributed among processing units and (2) how and what should each worker communicate.
This talk will introduce important principles in designing distributed machine learning algorithms as well as some recent developments in the field. Some examples using Apache Spark will be given.