IBM AI Systems Day       


IBM AI Systems Day - overview

IBM AI Systems Day 2018


The 2018 AI Systems Day will be held at the IBM Research Lab in 1 Rogers St, Cambridge, MA, on Wednesday, October 3th, 2018. This is a one day event that will feature a keynote, several short conference-style talks, and opportunities to discuss current research and results in systems and AI. It is part of the AI Research Week hosted by the MIT-IBM Watson AI Lab.

The event is open to all researchers and we encourage participation from students and professors. The main goal of the event is to increase awareness of each other's work, and to encourage interaction and collaboration.

AI Systems Day is held in cooperation with MIT as part of the MIT-IBM Watson AI Lab and is sponsored by the AI Systems reseach community.

Topics of interest are both Systems for AI and AI for systems. They include but are not limited to:

  • AI platforms
  • algorithm toolkits
  • AI programming languages
  • data structures
  • distributed learning
  • GPU processing
  • data visualization
  • AI lifecycle acceleration
  • AI application composition
  • automated ML and synthesis
  • HCI of AI
  • security and ethics
  • hardware for AI

Registration and Logistics:

Registration is free but required. To register pleas visit:

Invited Speaker:

Soumith Chintala (Facebook)
Usable while Performant: the challenges building PyTorch

With PyTorch, a deep learning framework focused on usability, we built software that is obsessively focused on usability and user’s debugging needs, while keeping performance within epsilon of other software offering the same functionality. In this talk, we shall go through some of the decisions made in designing PyTorch, and the constant tension between doing whole-graph optimizations to exploit obvious compilation wins, while keeping computations local, so that users don’t lose the ability to program in an interpreter.

Bio: Soumith Chintala is a researcher at Facebook AI Research. He helped build PyTorch, a popular deep learning framework and maintained It’s predecessor Torch-7. He also benchmarks and evaluates various hardware and platforms for deep learning workloads. In his previous life, he built mobile deep learning software for Android and iOS.


  • 09:00-09:25 - Welcome Breakfast
  • 09:25-10:30 - Session 1
    • TRIMS: Transparent and Isolated Model Sharing for Low Latency Deep Learning Inference in Function as a Service Environments (slides)
      Abdul Dakkak (UIUC), Cheng Li (UIUC), Jinjun Xiong (IBM Research), Wen-mei Hwu (UIUC)
    • Cache Optimized Model Serving with Resource Constraints (slides)
      Anthony Thomas (IBM Almaden/UCSD), Niketan Pansare (IBM Almaden), Berthold Reinwald (IBM Almaden)
    • Dynamic Data-Driven Learning for Self-Healing Avionics
      Shigeru Imai, Sida Chen, Wennan Zhu, Carlos A. Varela (RPI)
    • Dataflow Matrix Machines and V-values: a Bridge between Programs and Neural Nets (slides)
      Michael Bukatin (HERE Technologies), Jon Anthony (Boston College)
    • A Study on the Fragility of Clustering (slides)
      Xin Yin, Vincenzo Musco, Iulian Neamtiu (NJIT)
  • 10:30-11:00 - Coffee Break
  • 11:00-12:00 - Keynote
  • 12:00-13:00 - Lunch - Cafeteria Annex
  • 13:00-14:00 - Session 2
    • DBPal: A Learned Natural Language Interface for Databases using Distant Supervision (slides)
      Carsten Binnig (TU Darmstadt), Ugur Cetintemel (Brown University), Prasetya Utama (TU Darmstadt), and Nathaniel Weir (Brown University)
    • DeepPress: Deep Semantic Compression for Relational Data (slides)
      Amir Ilkhechi, Michael (Yicong) Mao, Alex Galakatos, Andrew Crotty, Ugur Cetintemel (Brown University)
    • ReJOIN: A Prototype Query Optimizer Using Deep Reinforcement Learning (slides)
      Ryan Marcus, Olga Papaemmanouil (Brandeis)
    • Mocking Atari for Deep Reinforcement Learning (slides)
      Emma Tosch (UMass Amherst), Akanksha Atrey (UMass Amherst), John Foley (Smith College), Sam Witty (UMass Amherst), Kaleigh Clary (UMass Amherst), David Jensen (UMass Amherst)
    • simple_rl: Lightweight Reinforcement Learning in Python (slides)
      David Abel (Brown University)
  • 14:00-14:30 - Coffee Break
  • 14:30-15:30 - Session 3
    • INCEPTIONN: A Network-Centric Hardware/Algorithm Co-Design to Accelerate Distributed Training of Deep Neural Networks
      Youjie Li (UIUC), Jongse Park (Georgia Tech), Mohammad Alian (UIUC), Yifan Yuan (UIUC), Zheng Qu (THU), Peitian Pan (SJTU), Ren Wang (Intel), Alexander Gerhard Schwing (UIUC), Hadi Esmaeilzadeh (UCSD), Nam Sung Kim (UIUC)
    • DNNBuilder: an Automated Tool for Building High-Performance DNN Hardware Accelerators for FPGAs (slides)
      Xiaofan Zhang (UIUC), Junsong Wang (IBM Research China), Chao Zhu (IBM Research China), Yonghua Lin (IBM Research China), Jinjun Xiong (IBM T. J. Watson Research Center), Wen-mei Hwu (UIUC), Deming Chen (UIUC)
    • Modeling and mitigating the effects of memory failures in deep neural networks
      Karthik Swaminathan, Nandhini Chandramoorthy, Martin Cochet, Karthikeyan Shanmugam, Amit Dhurandhar, Schuyler Eldridge, Rajiv Joshi, Alper Buyuktosunoglu, Pradip Bose (IBM Research)
    • Mock Your RNGs: Validating software with stochastic algorihtms (slides)
      Daniel Lee (Generable)
    • Evaluating 'Graphical Perception' with CNNs
      Daniel Haehn, James Tompkin, Hanspeter Pfister (Harvard)
  • 15:30-16:00 - Coffee Break
  • 16:00-17:00 - Session 4
    • Joint bottom-up/top-down machine learning structures to simulate human audition and musical creativity (slides)
      Jonas Braasch, (Rensselaer Polytechnic Institute)
    • Reagent: Converting Ordinary Webpages into Interactive Software Agents (slides)
      Matthew Peveler (RPI), Jeff Kephart (IBM)
    • Are you talking to me? - An attention-aware embodied agent (slides)
      Rahul R. Divekar (RPI), Jeff Kephart (IBM), Lisha Chen (RPI), Xiangyang Mou (RPI), Hui Su (IBM and RPI)
    • Building Empathy with Chatbots (slides)
      Justin Weisz, Ingrid Lange, J Johnson, Narendra Joshi, Mohit Jain (IBM)

Important Dates:

  • Abstract submission: 6 September 2018
  • Decision: 17 September 2018
  • Registration: 26 September 2018
  • IBM AI Systems Day: 3 October 2018

Call for Submissions:

We welcome all topics related to Systems and AI, and encourage all results, including preliminary work and progress reports. Presentation of work accepted or already appeared at some conference with proceedings are also welcome. Talks are about 10 minutes long. If you want to present your work, please submit a title and abstract (about 300 words).

Topics for presentations can be submitted on: close


Selection Committee:



We welcome everyone to arrive at the IBM Research Lab in Cambridge starting at 9:00AM.


IBM Research Lab
1 Rogers St
Cambridge, MA 02142


Organizing Committee:

Important dates

  • Abstract submission: 6 September 2018
  • Decision: 17 September 2018
  • Registration: 26 September 2018
  • IBM AI Systems Day: 3 October 2018