8am - 8:05am: Introduction to ParLearning 2019
8:05am - 9am: Keynote talk 1: Accelerating Deep Learning with Tensor Processing Units - Dr. Lifeng Nai (Google Research, Mountain View, CA, USA)

Abstract: Google's Tensor Processing Unit (TPU), first deployed in 2015, provides services today for more than one billion people and provides more than an order of magnitude improvement in performance and performance/Watt compared to contemporary platforms. Inspired by the success of the first TPU for neural network inference, Google has developed multiple generations of machine learning supercomputers for neural network training that allow near linear scaling of ML workloads running on TPUv2 and TPUv3. In this talk, we will present how TPU works as a machine learning supercomputer to benefit a growing number of Google services. We will have a deep dive into TPU’s system & chip architecture and our hardware/software codesign methodology that turns accelerator concepts into reality.

9am - 9:30am: Regular paper 1: Large Scale Cloud Deployment of Spectral Topic Modeling
9:30am - 10am: Coffee break
10am - 10:45am: Keynote talk 2: Bots, Socks and Vandals: Malicious Actors in Online Forums - Professor V.S. Subrahmanian (Dartmouth College, Hanover, NH, USA)

Abstract: Malicious actors are omnipresent in online social and crowdsourced platforms – vandals on Wikipedia, bots on Twitter, and trolls on various platforms all play a major role in degrading the quality of open information and free discussion on the web.This talk will focus on the role of semantics and its relationship with networks in order to classify users on Twitter as bots and users on Wikipedia as vandals. In the context of Twitter bots, this talk will discuss the DARPA Twitter Bot Challenge and subsequent research. In the context of Wikipedia, Professor will also discuss the vandal early warning system (VEWS) and its role in identifying vandals as early as possible. Time permitting, this talk will discuss malicious actors in other online networks such as Slashdot and/or on e-commerce sites such as Flipkart. The talk reflects joint work with many students and colleagues.

10:45am - 11:30am: Keynote talk 3: HW-SW Codesign for AI at Facebook - Dr. Satish Nadathur (Facebook Research, Menlo Park, CA, USA)

Abstract: In this talk, we will describe the principles of HW-SW codesign at Facebook. We will go over key characteristics of deep learning workloads from a HPC perspective – in terms of how limited they are by compute and memory bandwidth. Based on our growing application needs, we need specialized ASIC solutions to keep up with the workload scale. We will describe the overall building blocks of the Facebook-designed hardware released to the Open Compute Project (OCP) that efficiently helps us deploy hardware at scale. We will end by sketching some of the key SW challenges in order to best leverage these systems.

11:30am - 12pm: Regular paper 2: Expedite Neural Network Training via Software Techniques
12pm - 12:30pm: Regular paper 3: Scaling up Stochastic Gradient Descent for Non-convex Optimisation

Scaling up machine-learning (ML), data mining (DM) and reasoning algorithms from Artificial Intelligence (AI) for massive datasets is a major technical challenge in the time of "Big Data". The past ten years have seen the rise of multi-core and GPU based computing. In parallel and distributed computing, several frameworks such as OpenMP, OpenCL, and Spark continue to facilitate scaling up ML/DM/AI algorithms using higher levels of abstraction. We invite novel works that advance the trio-fields of ML/DM/AI through development of scalable algorithms or computing frameworks. Ideal submissions should describe methods for scaling up X using Y on Z, where potential choices for X, Y and Z are provided below.

Scaling up

• Recommender systems
• Optimization algorithms (gradient descent, Newton methods)
• Deep learning
• Distributed algorithms and AI for Blockchain
• Sampling/sketching techniques
• Clustering (agglomerative techniques, graph clustering, clustering heterogeneous data)
• Probabilistic inference (Bayesian networks)
• Graph algorithms, graph mining and knowledge graphs
• Graph neural networks
• Autoencoders and variational autoencoders
• Generative adversarial networks
• Generative models
• Deep reinforcement learning

Using

• Parallel architectures/frameworks (OpenMP, CUDA etc.)
• Distributed systems/frameworks (MPI, Spark, etc.)
• Machine learning frameworks (TensorFlow, PyTorch etc.)

On

• Various infrastructures, such as cloud, commodity clusters, GPUs, and emerging AI chips.

Best Paper Award: The program committee will nominate a paper for the Best Paper award. In past years, the Best Paper award included a cash prize. Stay tuned for this year!

Travel Awards: Students with accepted papers will get a chance to apply for a travel award. Please find details on the ACM KDD 2019 website.

• Paper submission: May 12, 2019 (Anywhere on Earth)
• Author notification: June 1, 2019
• Camera-ready version: Jun 8, 2019

All submissions are limited to a total of 6 pages, including all content and references, and must be in PDF format and formatted according to the new Standard ACM Conference Proceedings Template. Additional information about formatting and style files is available online as the ACM Master Article Template. Papers that do not meet the formatting requirements will be rejected without review.

All submissions must be uploaded electronically at EasyChair.

We are planning to publish a special issue of a journal, consisting of the best papers of ParLearning 2019. We are about to publish a special issue of the Springer journal Future Generation Computer Systems, containing the selected papers of ParLearning 2017.

• Professor V.S. Subrahmanian (Dartmouth College, Hanover, NH, USA)
• Dr. Lifeng Nai (Google, Mountain View, CA, USA)
• Dr. Satish Nadathur (Facebook Infrastructure, Menlo Park, CA, USA)

• General Chairs: Arindam Pal (TCS Research and Innovation, Kolkata, India) and Henri Bal (Vrije Universiteit, Amsterdam, Netherlands)
• Program Chairs: Azalia Mirhoseini (Google AI, Mountain View, CA, USA) and Thomas Parnell (IBM Research, Zurich, Switzerland)
• Publicity Chair: Anand Panangadan (California State University, Fullerton, CA, USA)
• Steering Committee Chairs: Sutanay Choudhury (Pacific Northwest National Laboratory, Richland, WA, USA) and Yinglong Xia (Facebook AI, Menlo Park, CA, USA)

The first 7 editions of ParLearning were organized in conjunction with the International Parallel and Distributed Processing Symposium (IPDPS). The details of the past workshops can be found on the old website of ParLearning. From 2019, the organizers have decided to conduct it with KDD.

• ParLearning 2018
• ParLearning 2017
• ParLearning 2016
• ParLearning 2015
• ParLearning 2014
• ParLearning 2013
• ParLearning 2012