Sunday, February 1, 2015

ML/NLP MEETUPS RECORDINGS

Bay Area NLP (Natural Language Processing)

Combining Human and Machine Intelligence for NLP
Data trumps algorithms for Natural Language Processing. Where the right machine learning algorithm can increase accuracy by 5%, annotating more training data often gives 20% or greater improvement. This can be the difference between 70% and 90% accuracy, and for many of Idibon’s clients this is the difference between unusable and actionable information. Rob will go over how Idibon approaches the combination of human and machine intelligence. Once humans are in the loop, they become the most expensive part of the equation. Increasing the accuracy efficiently is the complex co-optimization of both machine learning and analyst interaction, taking into account factors like: data coverage; the attention and agency of the analysts; the interfaces that allow analysts to annotate and supply better data; and strategies to resolve inter-analyst disagreement.

Jul 23, 2015 · 6:00 PM
SF Green Space

http://www.meetup.com/Bay-Area-NLP/events/223296301/

a recap at:
http://idibon.c



Computing with Affective Lexicons, Dan Jurafsky

Computing with
Affective Lexicons
Affective, Sentimental,
and Connotative
Meaning in the Lexicon
https://web.stanford.edu/~jurafsky/slp3/slides/21_SentLex.pdf




 Exotic Deep Learning Workshop
http://www.meetup.com/SF-Data-Science/events/226462698/

About the Workshop:

In this workshop, Adam Gibson will cover an overview of the deeplearning4j framework for practical applications of deep learning. We will cover basic terminology of deep neural networks applicable to the tuning of deep learning algorithms for applications ranging from:
Anomaly detection in video (Convolutional Net -> Recurrent)Classification (Multi layer perceptron)Generative textual models of shakespeare (recurrent nets).

What' You'll Takeaway:

The goal of this meetup will be to ensure learning of what components of a neural network algortihm are common to several tasks while also ensuring we understand the tradeoffs of different architectures targeted at different applications.

The core of the content will be a stress on tools as well as live coding of the above examples. The objective for attendees will be to walk away with a fundamental understanding of how to apply neural networks to time series data as well as basic classification.

Pre-requisites:


The expected attendee should have a basic understanding of machine learning and ideally have seen some basic neural networks before.

Java experience is preferred but not required for attendees.

Meet Your Instructor:
Adam Gibson, founder of Skymind.io

An Additional Workshop :)

After Adam's workshop, there will be a follow up workshop with Matthew Maloney, Co-founder of FCell, Founder of Earthquake Enterprises and Founder of Forensic Comparison Software. Details are TBD.

SVML 20150130 Predictive Models for User Behavior in Retail
http://youtu.be/Jf8HjCq43Dk


What you need to know about OAuth
for API Platform
http://www.meetup.com/Palo-Alto-Data-Science-Association/events/226260546/
https://www.youtube.com/watch?v=I_x38bwa-pk

Tsvi Achler: What is the brain doing different from machine learning algorithms?


Tsvi Achler has a unique background focusing on the neural mechanisms of recognition from a multidisciplinary perspective. He has done extensive work in theory and simulations, human cognitive experiments, animal neurophysiology experiments, and clinical training. He has an applied engineering background, has received bachelor degrees from UC Berkeley in Electrical Engineering, Computer Science and advanced degrees from University of Illinois at Urbana-Champaign in Neuroscience (PhD), Medicine (MD) and worked as a postdoc in Computer Science, and at Los Alamos National Labs, and IBM Research. He now heads his own startup Optimizing Mind whose goal is to provide the next generation of machine learning algorithms.
In his own words, below is an abstract of what Tsvi will talk to us about on Dec 9:
"The origin of phenomena observed in brain studies such as oscillations and a speed-accuracy tradeoff remain unclear.  It also remains unclear how the brain can be computationally flexible (quickly learn, modify, and use new patterns as it encounters them from the environment), and recall (reason with or describe recognizable patterns from memory).  I study the brain from multidisciplinary perspectives looking for a single, compact network that can display these phenomena and perform flexible recognition. 
Virtually all popular models of the brain and algorithms of machine learning remain “feedforward” even though it has been clear since the early days that this may limit flexibility (and is not optimal for recall, symbolic reasoning, or analysis).  Feedforward methods use optimized weights to perform recognition. In feedforward networks “uniqueness information” is encoded into weights based on the frequency of occurrence found in the training set.  This requires optimizing weights over the whole training set.
Instead, I suggest uniqueness is estimated during recognition, by performing optimization on the current pattern that is being recognized.  This is NOT optimization to learn weights, instead optimization to perform recognition.  Subsequently, only simple Hebbian-like relational learning is required during learning without any uniqueness information. The weights are no longer “feedforward” but learning is more flexible and can be much faster (>>100x), especially for big data since it does not require elaborate rehearsal.  From a phenomenological perspective, the optimization during recognition displays general properties observed in brain and cognitive experiments, predicting, oscillations, initial bursting with unrecognized patterns, and speed-accuracy tradeoff.
I will compare computational and cognitive properties of both approaches and discuss the state of new research initiatives."


Bay Area NLP Reading Group
http://www.meetup.com/Bay-Area-NLP/events/229771165/
To start, we will be following the winter 2016 schedule of Stanford NLP Reading Group.
For this session in particular, we'll be reading two papers by Jacob Andreas, Marcus Rohrbach, Trevor Darrell and Dan Klein:
• OPTIONAL BUT USEFUL BACKGROUND READING: An Introduction to Formal Computational Semantics


[Palo Alto] TensorFlow +Spark +Neural Nets +Deep Learning +Nvidia CUDA +OpenDeep

http://www.meetup.com/Advanced-Apache-Spark-Meetup/events/229132969/
Agenda 
Modularity in Neural Nets and Resulting Design Choices in Open Deep
(Markus Beissinger, Founder of Vitruvian Science Labs)
DeepLearning4J:  CUDA-based Deep Learning Java Library for Spark
(Adam Gibson, Founder of Skymind)
Spark Project Tungsten + GPUs:  Exploiting GPUs in Spark
(Kazuaki Ishizaki, Research Engineer, IBM Research - Tokyo)
Nvidia CUDA + GPUs + Spark:  Extending Spark Operators for Distributed Spark Matrix Multiplication in new, Row-grouped CSR Format for sparse matrices
(Maxim Naumov, Sr. Research Scientist @ Nvidia) 
SkFlow = Scikit-learn + TensorFlow + GPUs!! (Chris Fregly, Principal Data Solutions Engineer, IBM Spark Tech Center)
TensorFlow Serving:  Real-time Prediction Layer similar to Prediction.IO
(Chris Fregly, Principal Data Solutions Engineer, IBM Spark Tech Center)

Related Links


Boris Babenko of Orbital Insight will talk about Deep Learning and the Analysis of Satellite Imagery.


Advanced Spark and TensorFlow Meetup
http://www.meetup.com/Advanced-Apache-Spark-Meetup/
[advanced-spark-tensorflow] Videos from the recent Spark 2.0 + TensorFlow Meetup
http://www.meetup.com/Advanced-Spark-and-TensorFlow-Meetup/events/230506483/

Spark Workshop on the Peninsula
http://www.meetup.com/SF-Big-Analytics/events/229293237/

GraphFrames: DataFrame-based graphs for Apache Spark
http://www.meetup.com/spark-online/events/229846044/
Cloud AI Robotics Tech Talk - TensorFlow: Deep Learning for Everyone
http://www.meetup.com/Cloud-AI-Robotics/events/233322698/
Online live streaming is available:
http://www.youtube.com/channel/UCfRdSkz3eW7eU-AogaBXUiw/live

HOW TO BECOME A EXPERT ON APACHE SPARK AND SCALA
http://www.meetup.com/Bay-Area-Mesos-User-Group/events/234917936/
recording
http://unbouncepages.com/live-apachespark-webinar/

Speech and Conversational AI
http://www.meetup.com/Bay-Area-NLP/events/234919977/

Election Data Hackathon
https://www.meetup.com/Bay-Area-Women-in-Machine-Learning-and-Data-Science/events/234866195/

https://github.com/WiMLDS/election-data-hackathon

 https://www.kaggle.com/benhamner/clinton-trump-tweets
http://varianceexplained.org/r/trump-tweets/


Parvez Ahammad: SpeedPerception
http://www.meetup.com/SF-Web-Performance-Group/events/234840935/

• Dataset, study and results: http://speedperception.com/
• Challenge Site:http://speedperception.meteorapp.com/challenge

Data Science in Education Group -- San Francisco Bay Area
Talk. UC Berkeley Prof. Hearst on Natural Language Processing
https://www.meetup.com/degree/events/235084021/
CAN NATURAL LANGUAGE PROCESSING BECOME NATURAL LANGUAGE COACHING?
Speaker
Prof. Marti A. Hearst, UC Berkeley
Talk Summary
How we teach and learn is undergoing a revolution, due to changes in technology and connectivity. Education may be one of the best application areas for advanced NLP techniques, and NLP researchers have much to contribute to this problem, especially in the areas of learning to write, mastery learning, and peer learning. In this talk I consider what happens when we convert natural language processors into natural language coaches.

Speaker's bio

Marti Hearst is a Professor at UC Berkeley in the School of Information and EECS. She received her PhD in CS from UC Berkeley in 1994 and was a member of the research staff at Xerox PARC form 1994-1997. Her research is in computational linguistics, search user interfaces, information visualization, and improving learning at scale. Her NLP work includes automatic acquisition of hypernym relations ("Hearst Patterns"), TextTiling discourse segmentation, abbreviation recognition, and multiword semantic relations. She wrote the book "Search User Interfaces" (Cambridge) in 2009, co-founded the ACM Conference on Learning at Scale in 2014, is Vice President Elect of the Association for Computational Linguistics and was named an ACM Fellow in 2013. She has received four student-initiated Excellence in Teaching Awards.

Silicon Valley Big Data Science
Competitive Data Science - Brain dump from Kaggle Grand Masters(s)
https://www.meetup.com/Silicon-Valley-Big-Data-Science/events/236800410/


7:00 – 7:30 PM - Dmitry Larko: (Kaggle Grand Master)

Winning solution for the Grupo Bimbo Inventory Demand competition. 1st place out of 1969

https://www.kaggle.com/c/grupo-bimbo-inventory-demand

7:30 – 8:00 PM - Vladimir Iglovikov: (Kaggle Master)

Top 2% (40 out of 3055) solution for the Kaggle Allstate Claims Severity regression problem

https://www.kaggle.com/c/allstate-claims-severity

https://www.meetup.com/sfmachinelearning/

AI Meetup with Coline Devin, from UC Berkeley. She will be talking about her recent work on robot perception.
Bio: Coline is a PhD student in BAIR at UC Berkeley, advised by Professors Pieter Abbeel and Trevor Darrell. She graduated in Computer Science from Harvey Mudd College in May 2015, and has previously interned at Google Deepmind and MILA.
Talk Title: Deep Object-Centric Representations for Generalizable Robot Learning
The talk will be primarily centered around this paper: https://arxiv.org/pdf/1708.04225.pdf
Talk Abstract:
Robotic manipulation in complex open-world scenarios requires both reliable physical manipulation skills and effective and generalizable perception. In this paper, we propose a method where general purpose pretrained visual models serve as an object-centric prior for the perception system of a learned policy. We devise an object-level attentional mechanism that can be used to determine relevant objects from a few trajectories or demonstrations, and then immediately incorporate those objects into a learned policy. A task-independent meta-attention locates possible objects in the scene, and a task-specific attention identifies which objects are predictive of the trajectories. The scope of the task-specific attention is easily adjusted by showing demonstrations with distractor objects or with diverse relevant objects. Our results indicate that this approach exhibits good generalization across object instances using very few samples, and can be used to learn a variety of manipulation tasks using reinforcement learning.































































Posted by at

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)