Welcome to the 4th IEEE BDTL!

-- Heterogeneous Representation and Networks

Although widely applied on considerable scientific research, conventional statistical machine learning revolves on a simplified assumption that the training data, from which the algorithms learn, are drawn i.i.d. from the same distribution as the test data, to which the learned models are applied. This assumption, being broken down by numerous real-world applications nowadays, especially with the emergence of large-scale data from the private internal data, or the public Internet, has fundamentally restricted the development of practical learning algorithms. More examples include but are not limited to: (1) speech recognition when speakers have strong dialects from different countries, regions, genders, or aging groups; (2) surveillance system where captured suspects' faces are side view, under unknown lighting conditions, in low-resolutions, different from the conditions of the registered faces in the system.

We may attribute above issues to heterogeneous feature representation of different data/object types without explicitly modeling the connections between data samples. This is also the research target of recently developed transfer learning (TL) that bridges the gap between the feature space of heterogeneous concepts. When it comes to the structured data, e.g., social networks, academic networks, however, a joint modeling of both data attributes/features, and their interaction represented by networks becomes necessary and feasible. This is exactly the motivation of recent research in the field of heterogeneous information networks (HIN).

It is thus our intention to promote conventional transfer learning in vectorized feature space and extend to heterogenous information network, where each object is characterized by both object attributes and link types with other objects. This aligns well with most existing real-world information networks and their learning tasks, and becomes particularly useful under the context of “Big Data”, where knowledge transfer and reuse of well-established or labeled objects could help other linked objects in the variety of machine learning tasks. This half-day workshop (in conjunction with IEEE Big Data 2019) is a continuation of our past Big Data Transfer Learning (BDTL) workshops (1st BDTL2nd BDTL3rd BDTL) which will provide a focused international forum to bring together researchers and research groups to review the status of transfer learning on both conventional vectorized features and heterogeneous information networks. Specifically, we will discuss the challenges given large-scale networks, new deep computing models, and explore future directions particularly in the unconstrained social environments, such as social media data in the cloud, Facebook and YouTube applications. The workshop will consist of one to two invited talks together with peer-reviewed regular papers (oral and poster).

This half-day workshop (in conjunction with IEEE Big Data 2019) is a continuation of our past Big Data Transfer Learning (BDTL) workshops (1st BDTL, 2nd BDTL, 3rd BDTL) which will provide a focused international forum to bring together researchers and research groups to review the status of transfer learning on both conventional vectorized features and heterogeneous information networks. Specifically, we will discuss the challenges given large-scale networks, new deep computing models, and explore future directions particularly in the unconstrained social environments, such as social media data in the cloud, Facebook and YouTube applications. The workshop will consist of one to two invited talks together with peer-reviewed regular papers (oral and poster).

Original high-quality papers are solicited on a wide range of topics including:

  • New perspective, concepts, or theories on transfer learning, or domain adaptation
  • Web-scale transfer learning from loosely organized source to unlabeled target data
  • Novel applications of transfer learning on networked data
  • Real-time knowledge transfer for large-scale social stream
  • Large-scale transfer learning with latent, complex, or noisy source domain
  • Transferrable representation learning on heterogeneous information networks
  • Cross domain similarity/metric learning on heterogeneous information networks
  • Knowledge fusion on heterogeneous information networks
  • New datasets, benchmarks, and open-source software for transfer learning and heterogeneous information networks
  • Submission Deadline: Nov. 1st, 2019
  • Notification: Nov 10th, 2019
  • Camera Ready: Nov 20th, 2019
  • Workshop Date: Dec 9th, 2019, Afternoon Session
  • Short papers: 6 pages
  • Long papers: 10 pages
  • Latex template: latex
  • Word template: word
  • Paper click here to submit your paper online.

General Chairs

Workshop Co-Chairs

  • Ming Shao, University of Massachusetts Dartmouth, USA

Tentative Program Committee

  • Raghuraman Gopalan, Principal Scientist at AT&T Labs-Research, USA
  • Boqing Gong, Google Research, USA
  • Mingming Gong, The University of Melbourne, Australia
  • ChunWei Sean, Technical Staff at DSO National Laboratories, Singapore
  • Neela Rahimi, UMass Dartmouth
  • Riazat Ryan, UMass Dartmouth

Web and Publicity Co-Chairs

  • Deepak Kumar, University of Massachusetts Dartmouth, USA

Date: Dec 9th, 2019

Location: Mt Washington, The Westin Bonaventure Hotel & Suites, Los Angeles

Workshop Schedule:

1:30 PM--2:00 PM Opening Remarks Workshop co-chair
2:00 PM--2:25 PM Decoder Transfer Learning for Predicting Personal Exposure to Air Pollution PEIJIANG ZHAO and Koji Zettsu
2:25 PM--2:50 PM Advertiser-Assisted Behavioral Ad-Targeting via Denoised Distribution Induction Kei Yonekawa, Niu Hao, Mori Kurokawa, Arei Kobayashi, Daichi Amagata, Takuya Maekawa, and Takahiro Hara
2:50 PM--3:15 PM DC2: A Divide-and-conquer Algorithm for Large-scale Kernel Learning with Application to Clustering Ke Alexander Wang, Xinran Bian, Pan Liu, and Donghui Yan
3:15 PM--3:40 PM On Online Hate Speech Detection. Effects of Negated Data Construction Mourad Oussalah and Abderraouf Cheniki
3:40--4:10: Coffee Hours    
4:10 PM--4:35 PM Multi-View, Generative, Transfer Learning for Distributed Time Series Classification Sreyasee Das Bhattacharjee, William J. Tolone, Ashish Mahabal, Mohammed Elshambakey, Isaac Cho, Abdullah al-Raihan Nayeem, Junsong Yuan, and George Djorgovski,
4:35 PM--4:40 PM Summary and Best Paper Award Workshop co-chair

The first workshop with this name was held in 2016, in conjunction with 2016 IEEE International Conference on Big Data (IEEE BigData 2016). So far, it has been successfully held 3 times. The homepages of previous two BDTL are as follows: