AAAI-23 Tutorial and Lab Forum

Schedule

The following list of tutorials and labs have been accepted for presentation at AAAI-23. All times listed are Eastern Standard Time (Washington, DC)

Tuesday, February 7, 2023

MORNING QUARTER-DAY TUTORIALS

8:30 am – 10:15 am

• TFQA1: Cooperative Multi-Agent Learning: A Review of Progress and Challenges (Room 204C)
  Yali Du, Joel Z Leibo

10:45 am – 12:30 pm

• TFQA2: Introducing Neuronal Diversity into Deep Learning (Room 204C)
  Feng-Lei Fan, Fei Wang

MORNING QUARTER-DAY LABS

8:30 am – 10:15 am

• LFQA1: Hands-On with the BLACK Satisfiability Checker (Room 204B)
  Luca Geatti, Nicola Gigante

10:45 am – 12:30 pm

• LFQA2: KGTK: User-Friendly Toolkit for Manipulation of Large Knowledge Graphs (Room 204B)
  Filip Illievski, Jay Pujara, Ke-Thia Yao, Gleb Satyukov, Kian Ahrabian

MORNING HALF-DAY TUTORIALS

8:30 am – 12:30 pm

• TFHA1: Trustworthy and Responsible AI: Fairness, Interpretability, Transparency and Their Interactions (Room 201)
  Yulin Zhou, Harsha Nori, Besmira Nushi, Jieyu Zhao, Leilani Gilpin
• TFHA2: The Polynomial Nets in Deep Learning Architecture (Room 203A)
  Grigorios Chrysos, Markos Georgopoulos, Razvan Pascanu, Volkan Cevher
• TFHA4: Machine Learning for Causal Inference (Room 202A)
  Zhixuan Chu, Jing Ma, Jundong Li, Sheng Li
• TFHA5: AI Fairness through Robustness (Room 206)
  Mikhail Yuochkin, Yuekai Sun, Pin-Yu Chen
• TFHA6: Advances in Neuro Symbolic Reasoning (Room 202B)
  Chitta Baral, Paulo Shakarian, Gerardo I. Simari, Alvaro Velasquez

AFTERNOON QUARTER-DAY TUTORIALS

2:00 pm – 3:45 pm

• TFQP1: Specification-Guided Reinforcement Learning (Room 205)
  Kishor Jothimurugan, Suguman Bansal, Osbert Bastani, Rajeev Alur

AFTERNOON QUARTER-DAY LABS

2:00 pm – 3:45 pm

• LFQP1: Time Series Anomaly Detection Tool: Hands of Lab (Room 204B)
  Dhaval Patel

AFTERNOON HALF-DAY TUTORIALS

2:00 pm – 6:00 pm

• TFHP1: Pervasive AI (Room 203A)
  Davide Bacciu, Antonio Carta, Patrizio Dazzi, Claudio Gallicchio
• TFHP2: Bi-level Optimization in Machine Learning: Foundations and Applications (Room 206)
  Sijia Liu, Mingyi Hong, Yihua Zhang, Bingqing Song
• TFHP3: Risk-Sensitive Reinforcement Learning via Policy Gradient Search (Room 204A)
  Prashanth L.A. and Michael Fu
• TFHP4: On Explainable AI: From Theory to Motivation, Industrial Applications, XAI Coding &; Engineering Practices (Room 201)
  Freddy Lecue, Fosca Giannotti, Riccardo Guidotti and Pasquale Minervini
• TFHP5: Everything You Need to Know about Transformers: Architectures, Optimization, Applications, and Interpretation (Room 202A)
  Andy Zeng, Boqing Gong, Chen Sun, Ellie Pavlick, Neil Houlsby
• TFHP6: Large-Scale Deep Learning Optimization Techniques (Room 202B)
  James Demmel, Yang You
• TFHP7: Inductive Logic Programming: An Introduction and Recent Advances (Room 204C)
  Andrew Cropper, Celine Hocquette, Sebastian Dumancic

Wednesday, February 8, 2023

MORNING HALF-DAY TUTORIALS

8:30 am – 12:30 pm

• TSHA1: Generalizable Commonsense Reasoning (Room 202B)
  Filip Ilievski, Alessandro Oltramari, Kaixin Ma, Peifeng Wang, Jonathan Francis
• TSHA2: Graph Neural Networks: Foundation, Frontiers and Applications (Room 202A)
  Lingfei Wu, Peng Cui
• TSHA3: The Economics of Data and Machine Learning (Room 209BC)
  Haifeng Xu, James Zou, Shuran Zeng
• TSHA4: Never-Ending Learning, Lifelong Learning and Continual Learning: Systems, Models, Current Challenges and Applications (Room 201)
  Estevam Hruschka

MORNING HALF-DAY LABS

8:30 am – 12:30 pm

• LSHA1: Subset Selection in Machine Learning: Hands-On Application with CORDS, DISTIL, SUBMODLIB, and TRUST (Room 203A)
  Nathan Beck, Suraj Kothawade, Krishnateja Killamsetty, Rishabh Iyer
• LSHA2: Automated AI For Decision Optimization with Reinforcement Learning (Room 204C)
  Dharmashankar Subramanian, Takayuki Osogami, Radu Marinescu, Alexander Zadorojniy, Long Vu, Nhan H. Pham
• LSHA3: Colossal-AI: Scaling AI Models in Big Model Era (Room 204A)
  James Demme, Yang Yo
• LSHA4: Building Approachable, Hands-On Embedded Machine Learning Curriculum Using Edge Impulse and Arduino (Room 206)
  Brian Plancher, Shawn Hymel
• LSHA5: OpenMMLab: A Foundational Platform for Computer Vision Research and Production (Room 204B)
  Kai Chen, Ruohui Wang, Songyang Zhang, Wenwei Zhang, Yanhong Zeng
•  

AFTERNOON QUARTER-DAY TUTORIALS

2:00 pm – 3:45 pm

• TSQP1: Data Compression with Machine Learning (Room 202B)
  Karen Ulrich, Yibo Yang, Stephan Mandt

4:15 pm – 6:00 pm

• TSQP2: Towards Causal Foundations of Safe AI (Room 202B)
  Tom Everitt, Lewis Hammond, Jon Richens

AFTERNOON HALF-DAY TUTORIALS

2:00 pm – 6:00 pm

• TSHP1: Hyperbolic Neural Networks: Theory, Architectures and Applications (Room 204B)
  Nurendra Choudhary, Karthik Subbian, Srinivasan H. Sengamedu, Chandan K. Reddy
• TSHP2: AI for Data-Centric Epidemic Forecasting (Room 203A)
  Alexander Rogriguez, Harshavardhan Kamarthi, B. Aditya Prakash
• TSHP3: Recent Advances in Bayesian Optimization (Room 202A)
  Jana Doppa, Aryan Deshwal, Syrine Belakaria
• TSHP4: Teaching Computing and Technology Ethics: Engaging Through Science Fiction (Room 204A)
  Emanuelle Burton, Judy Goldsmith, Nicholas Mattei, Cory Siler, Sara-Jo Swiatek
• TSHP5: Knowledge-Driven Vision-Language Pretraining (Room 201)
  Manling Li, Xudong Lin, Jie Lei, Mohit Bansal, Shih-Fu Chang, Heng Ji

AFTERNOON HALF-DAY LABS

2:00 pm – 6:00 pm

• LSHP1:Optimization with Constraint Learning (Room 206)
  Ilker Birgil, Donato Maragno, Orit Davidovich
• LSHP2: Automated Machine Learning & Tuning with FLAML (Room 209BC)
  Chi Wang, Qingyun Wu, Xueqing, Luis Quintanilla
• LSHP3: Innovative Uses of Synthetic Data (Room 204C)
  Mihaela van der Schaar, Zhaozhi Qian

TFQA2: Introducing Neuronal Diversity into Deep Learning
Feng-Lei Fan, Fei Wang

The brain is the most intelligent system we have ever known so far. Given the incredible capability of the human brain, neuroscience has been constantly supporting deep learning as a think tank and a validation tool, e.g., inspiring the invention of neocognitron, the pioneering work of convolutional models. Although a brain remains vastly undiscovered, it is clear that the existing deep learning still goes far behind the human brain in many important aspects such as efficiency, interpretability, memory, and robustness. Currently, the characteristics of the mainstream deep learning models are fundamentally different from the biological neural system in terms of neuronal diversity. While deep learning models are typically built on a single universal primitive neuron type, the human brain has numerous morphologically and functionally different neurons. It is no exaggeration to say that neuronal diversity is an enabling factor for all kinds of biological intelligent behaviors. Since an artificial network is a miniature of the human brain, introducing neuronal diversity should be able to add value in terms of addressing the essential problems of artificial networks. There have been increasingly many studies dedicated to neuronal diversity in deep learning, such as polynomial neurons and pyramidal neurons. This tutorial will summarize them in detail. In particular, we will also focus on how neuronal diversity can benefit deep learning in four aspects: efficiency, memory, interpretability, and robustness. We will also envision how neuronal diversity pushes the boundary of deep learning and its potential of escalating deep learning into a new stage.

LFQA2: KGTK: User-Friendly Toolkit for Manipulation of Large Knowledge Graphs
Filip Illievski, Jay Pujara, Ke-Thia Yao, Gleb Satyukov, Kian Ahrabian

Typical use cases of working with knowledge graphs require a host of different approaches, tools, and formats to obtain, manipulate, and customize the data. Composing such a pipeline today requires knowledge of multiple tools that are not designed to work together, making it costly to implement these pipelines for large KGs and hindering massive adoption of knowledge graphs by AI developers and researchers. In this tutorial, we will present KGTK: an integrated Knowledge Graph ToolKit, which provides a wide range of knowledge graph manipulation and analysis functionality and supports common use cases such as large-scale network analysis, data enrichment, and quality estimation. The lab will showcase four diverse applications of KGTK on 1) analyzing publication graphs and leveraging them to drive scientific innovation; 2) supply chain and financial transaction analytics; 3) event analytics with moral dimensions; and 4) commonsense knowledge harmonization and reasoning. Each session will be based on a jupyter notebook in Python, available both on Google Colab as well as for local installation.

TFHA2: The Polynomial Nets in Deep Learning Architectures
Grigorios Chrysos, Markos Georgopoulos, Razvan Pascanu, Volkan Cevher

Polynomial networks enable a new network design that treats a network as a high-degree polynomial expansion of the input. Polynomial networks have demonstrated state-of-the-art performance in a range of tasks. Despite the fact that polynomial networks have appeared for several decades in machine learning and complex systems, they are not widely acknowledged for their role in modern deep learning. In this tutorial we intend to bridge the gap and draw parallelisms between modern deep learning approaches and polynomial networks.  To this end, we will share recent developments on the topic, as well as explain the required tools.

TFHA4: Machine Learning for Causal Inference
Zhixuan Chu, Jing Ma, Jundong Li, Sheng Li

Causal inference has numerous real-world applications in many domains such as education, health care, political science, marketing, and online advertising. Causal inference has been studied for decades; however, traditional methods may have limited capability on handling large-scale and high-dimensional heterogeneous data. Recent research efforts demonstrate that machine learning could greatly facilitate causal inference tasks such as treatment effect estimation and counterfactual inference. Meanwhile, casual knowledge extracted from observational data could enhance the reliability of machine learning models. In this tutorial, we will present a comprehensive overview of the intersection of causal inference and machine learning. We will start with the background of causal inference and briefly introduce several traditional causal inference methods. Then we will introduce the state-of-the-art machine learning algorithms for causal inference, especially the representation learning based methods and graph neural networks-based methods. In addition, we will discuss how to exploit causal knowledge for reliable machine learning and showcase promising applications of these methods in multiple domains.

TFHA6: Advances in Neuro Symbolic Reasoning
Chitta Baral, Paulo Shakarian, Gerardo I. Simari, Alvaro Velasquez

Over the past five years, the community has made significant advances in neuro symbolic reasoning (NSR).  These NSR frameworks are now capable of embedding prior knowledge in deep learning architectures, guiding the learning process with logical constraints, providing symbolic explainability, and using gradient-based approaches to learn logical statements.  At this time, several approaches are seeing usage in various application areas.  This tutorial is designed for researchers looking to understand the current landscape of NSR research as well as those looking to apply NSR research in areas such as natural language processing and verification.  The pace of progress in NSR is expected to continue as firms such as IBM, Samsung, and Lockheed-Martin are now heavily investing in research in this area inaddition to the recently announced government-funded efforts such as DARPA’s ASNR program indicate that this area will grow. A secondary goal of this tutorial is to help build a larger community around this topic as more basic researchers and applied scientists turn to NSR to build upon the successes of deep learning.  Attendees of the tutorial should be familiar with concepts in deep learning and logical reasoning, have mathematical maturity, as well as a basic understanding of fuzzy/real-valued logic.

Additionally, the following students will be supporting the tutorial for creation of materials:

LFQP1: Time Series Anomaly Detection Tool: Hands of Lab
Dhaval Patel

Asset Health and Monitoring is an emerging AI Application that aims to deliver efficient AI-powered solutions to various industrial problems such as anomaly detection, failure  pattern analysis, etc. In this lab-based tutorial, we present a web-based time series anomaly detection tool – a new scikit-learn compatible toolkit specialized for the time series-based anomaly detection problem. The key focus of our tutorial includes the design and development of an anomaly detection pipeline, a zero-configuration interface for automated discovery of an anomaly pipeline for any given dataset (univariate and multi-variate), a set of 5 frequently used workflow empirically derived from past experiences, a scalable technique for conducting efficient pipeline execution. We extensively tested deployed anomaly detection services using multiple datasets with varying time-series data characteristics.

TFHP2: Bi-level Optimization in Machine Learning: Foundations and Applications
Sijia Liu, Mingyi Hong, Yihua Zhang, Bingqing Song

Bi-level machine learning has become an emerging but overlooked topic that leverages bi-level optimization (BLO) to tackle a wide class of artificial intelligence challenges such as: robust AI, efficient AI, and generalizable AI. Our half-day tutorial will systematically review various aspects of BLO, including theoretical foundations, algorithm development, and practical applications. Specifically, the content will cover detailed discussions on recent advances in the theoretical developments of BLO, its applications in different fields of AI, as well as its efficient implementation. In addition to technical presentations, we will also offer a carefully designed Demo Expo to showcase the practical implementations of BLO methods and their applications. The expo contains an in-person implementation tutorial, an actively maintained toolbox for BLO in GitHub, and a regularly updated blog channel to facilitate virtual participation during and after the conference.

TFHP4: On Explainable AI: From Theory to Motivation, Industrial Applications, XAI Coding & Engineering
Practices
Freddy Lecue, Fosca Giannotti, Riccardo Guidotti and Pasquale Minervini

The future of AI lies in enabling people to collaborate with machines to solve complex problems. Like any efficient collaboration, this requires good communication, trust, clarity and understanding. XAI (eXplainable AI) aims at addressing such challenges by combining the best of symbolic AI and traditional Machine Learning. Such a topic has been studied for years by all different communities of AI, with different definitions, evaluation metrics, motivations and results. This tutorial is a snapshot on the work of XAI to date, and surveys the work achieved by the AI community with a focus on machine learning and symbolic AI related approaches (given the half-day format). We will motivate the needs of XAI in real-world and large-scale application, while presenting state-of-the-art techniques, with best XAI coding and engineering practices. In the first part of the tutorial, we give an introduction to the different aspects of explanations in AI. We then focus the tutorial on two specific approaches: (i) XAI using machine learning and (ii) XAI using a combination of graph-based knowledge representation and machine learning. For both we get into the specifics of the approach, the state of the art and the research challenges for the next steps. The final part of the tutorial gives an overview of real-world applications of XAI as well as best XAI coding and engineering practices, as XAI technologies are required to be seamlessly integrated in AI applications.

TFHP6: Large-Scale Deep Learning Optimization Techniques
James Demmel, Yang You

Large transformer models display promising performance on a wide spectrum of AI applications. However, there has been a recent insurgence of extremely large models due to their good performance. These models have exorbitant training costs due to large communication overhead and the number of computations they execute. Therefore, both academia and industry are scaling DL training on larger clusters. However, degrading generalization performance, non-negligible communication overhead, and increasing model size prevent DL researchers and engineers from exploring large-scale AI models. In this tutorial, we aim to provide a clear sketch of the optimizations for large-scale deep learning with regard to model accuracy and model efficiency. We investigate algorithms that are most commonly used for optimizing: we recall the key idea of gradient descent optimization, introduce large batch training optimization, elaborate on the debatable topic of the generalization gap that arises in large-batch training, present second-order optimization, and lastly, review the state-of-the-art strategies in addressing the communication overhead and reducing memory footprints. 

The tutorial will be self-contained and only a basic understanding of deep learning would be good enough to follow most materials. More details can be found at https://github.com/hpcaitech/ColossalAI/tree/main/colossalai/nn/optimizer.

TFHP7: Inductive Logic Programming: An Introduction and Recent Advances
Andrew Cropper, Celine Hocquette, Sebastian Dumancic

The main limitations of standard machine learning approaches include poor generalisation, a lack of interpretability, and the need for many training examples. However, unbeknown to many researchers, recent work in Inductive Logic Programming (ILP), a form of machine learning based on computational logic, has shown promise at addressing these limitations. This tutorial introduces ILP to a general AI audience by describing how to build an ILP system. We also discuss applications of ILP, such as game playing, drug design, and program synthesis. We highlight the main challenges and outline opportunities for collaboration with other AI areas, principally constraint solving.

This tutorial does not require any prerequisite knowledge but familiarity with first-order logic is helpful.

TSHA2: Graph Neural Networks: Foundation, Frontiers and Applications
Lingfei Wu, Peng Cui

The field of graph neural networks (GNNs) has seen rapid and incredible strides over recent years. Graph neural networks have become one of the fastest-growing research topics in machine learning, especially deep learning. This wave of research at the intersection of graph theory and deep learning has also influenced other fields of science, including recommendation systems, computer vision, natural language processing, inductive logic programming, program synthesis, software mining, automated planning, cybersecurity, and intelligent transportation.

This tutorial of Graph Neural Networks (GNNs): Foundation, Frontiers and Applications will cover a broad range of topics in graph neural networks, by reviewing and introducing the fundamental concepts and algorithms of GNNs, new research frontiers of GNNs, and broad and emerging applications with GNNs. In addition, rich tutorial materials will be included and introduced to help the audience gain a  systematic understanding by using our recently published book-Graph Neural Networks (GNN): Foundation, Frontiers and Applications, one of the most comprehensive books for researchers and practitioners for reading and studying in GNNs 

The audience is expected to have some basic understanding of data mining, machine  learning, linear algebra, and optimization. However, the tutorial will be presented at  college junior/senior level and should be comfortably followed by academic  researchers and practitioners from the industry.

TSHA4: Never-Ending Learning, Lifelong Learning and Continual Learning: Systems, Models, Current Challenges and Applications
Estevam Hruschka

In the last decades, different Never-Ending Learning (NEL) approaches have been proposed and applied in different tasks and domains. In addition, the variety of different names (never-ending learning, continual learning, lifelong learning, etc.), different assumptions (single/multi-model, neural/symbolic/hybrid, external/internal memory, etc.) used to describe systems and models that can keep learning in a continuous way, as well as the new achievements on self-supervised learning and multi-task learning (which are closely related to some of the NEL principles) present in large pre-trained models motivate the proposal of this tutorial. In summary, this tutorial aims at enabling the attendees to:

• Learn about the current state-of-the art on Never-ending Learning, Lifelong Learning and Continual Learning (and their variants);
• Understand the similarities and differences of the NEL principles and other approaches such as large pre-trained models, reinforcement learning and semi-supervised learning.
• Learn (guided by practical examples) how to model problems/tasks using NEL principles.
• Be prepared to follow along NEL principles and propose new approaches to boost performance of different AI applications.

Link to the tutorial webpage: https://megagon.ai/nel_tutorial/

LSHA2: Automated AI For Decision Optimization with Reinforcement Learning
Dharmashankar Subramanian,Takayuki Osogami, Radu Marinescu, Alexander Zadorojniy, Long Vu, Nhan H. Pham

Reinforcement learning has many practical real-world applications. However, RL solutions are highly sensitive to the choice of the RL algorithm and its internal hyperparameters which requires expert manual effort. This limits the widespread applicability of RL solutions in practice. In this lab session, we introduce an automated system, AutoDO, for end-to-end solving of sequential decision-making problems. Our system automatically selects the best RL algorithm and its hyperparameters using a search strategy based on limited discrepancy search coupled with Bayesian optimization. It supports both online and offline RL as well as automated Markov Decision Process (MDP) models based on mathematical programming.

Data scientists with very little or no experience in RL will be able to formulate and solve sequential decision-making problems using AutoDO. They will focus on preparing the inputs, namely a system environment for online RL and a data set annotated with a knowledge data structure for offline RL and automated MDP models and will gain experience in using our system to automate the solution pipeline generation for the optimal decision policy. Prerequisites will be shared in advance: mainly, basic familiarity with python and free subscription in advance to a public facing API service.

LSHA4: Building Approachable, Hands-On Embedded Machine Learning Curriculum Using Edge Impulse
and Arduino
Brian Plancher, Shawn Hymel

Embedded machine learning (ML) is the process of running machine learning algorithms on low-cost, resource-constrained microcontrollers and single board computers and is currently used to solve unique problems in industry and academia. As a result, there is a growing demand to teach embedded ML in higher education institutions to prepare developers, engineers, and researchers for these emerging data-driven approaches.

This lab provides hands-on experience creating an end-to-end embedded ML system using a combination of Edge Impulse and Arduino. In the lab, we will create and deploy a full image classification system by performing data collection using Arduino, model training with Edge Impulse, and live inference with Arduino. This will provide researchers and educators with the knowledge and tools to develop an approachable curriculum around embedded ML.  

Attendees should have an interest in finding ways to make ML easier and approachable to students, especially those outside of the computer science field. Such attendees could be professors or lecturers from other disciplines or those teaching ML in cross-cutting programs. Some programming experience will be needed (ideally C/C++ or Arduino). No machine learning experience is required.

TSQP1: Data Compression with Machine Learning
Karen Ulrich, Yibo Yang, Stephan Mandt

The efficient communication of information is of enormous societal and environmental impact and stands to benefit from the machine learning revolution seen in other fields. This tutorial disseminates the ideas from information theory and learned (neural) data compression to a broader machine learning and AI audience.

We focus on the core principles and techniques behind neural lossless and lossy compression and highlight the intimate connections between compression and statistical modeling, drawing upon deep generative models such as autoregressive models, VAEs, and GANs. These techniques apply to various problems beyond the standard data compression task, such as compression for perceptual quality and gradient compression in federated learning. We will also provide broader perspectives on the interplay between compression and machine learning, as well as the various challenges and societal issues of neural compression, such as fairness, trustworthiness, and energy consumption. We expect our audience to have basic knowledge of generative models such as autoregressive models, GANs, and especially latent variable models.

TSHP1: Hyperbolic Neural Networks: Theory, Architectures and Applications
Nurendra Choudhary, Karthik Subbian, Srinivasan H. Sengamedu, Chandan K. Reddy

While preliminary research in the domain of graph analysis was driven by neural architectures, recent studies have revealed important properties unique to graph datasets such as hierarchies and global structures. This has driven research into hyperbolic space due to their ability to effectively encode the inherent hierarchy present in graph datasets. The research has also been subsequently applied to other domains such as NLP and computer vision to get formidable results. However, the significant challenge for further growth is the obscurity of hyperbolic networks and a better comprehension of the necessary algebraic operations needed to broaden the application to different neural network architectures.

In this tutorial, we aim to introduce researchers and practitioners in the web domain to the hyperbolic equivariant of the Euclidean operations that are necessary to tackle their application to neural network architectures. Additionally, we describe the popular hyperbolic variants of GNN architectures such as recurrent networks, convolution networks, and attention networks, and explain their implementation, in contrast to the Euclidean counterparts. Furthermore, we also motivate our tutorial through existing applications in graph analysis, knowledge graph reasoning, product search, NLP, and computer vision and compare the performance gains to the Euclidean counterparts.

Prerequisites: Familiarity with popular neural network architectures such as MLP, RNNs, CNNs, and Attention networks.

TSHP3: Recent Advances in Bayesian Optimization
Jana Doppa, Aryan Deshwal, Syrine Belakaria

Many engineering and scientific applications including automated machine learning (e.g., neural architecture search and hyper-parameter tuning) involve making design choices to optimize one or more expensive to evaluate objectives. Bayesian Optimization (BO) is an effective framework to solve black-box optimization problems with expensive function evaluations. The goal of our tutorial is to present a comprehensive survey of BO starting from foundations to these recent advances by focusing onchallenges, principles, and algorithmic ideas and their connections. 

The target audience of this tutorial includes (1) General AI researchers and graduate students who will learn about principles, algorithms, and outstanding challenges to explore the frontiers of BO and its real-world applications; (2) BO researchers who will learn about the complete landscape of BO problems to gain breadth and learn about outstanding challenges in the frontiers of BO; (3) Industrial AI researchers and practitioners who will be apply the learned knowledge for solving automated machine learning and A/B testing problems; and (4) Researchers and practitioners working on science and engineering applications such as drug/vaccine design, materials design etc. will learn about useful BO tools.

TSHP5: Knowledge-Driven Vision-Language Pretraining
Manling Li, Xudong Lin, Jie Lei, Mohit Bansal, Shih-Fu Chang, Heng Ji

This tutorial targets AI researchers and practitioners interested in multimedia data understanding, such as text, images, and videos. Recent advances in vision-language pretraining connect vision and text through multiple levels of knowledge, including entity (object) knowledge, relation (scene graph) knowledge, event (activity) knowledge, procedural knowledge, and the knowledge from language models. These include methods to explicitly discover and encode structured knowledge extracted from text and vision data, and to retain semantic knowledge in language models. This tutorial will provide audience with a systematic introduction of (i) various vision-language pretraining methods for automated extraction, conceptualization and prediction of entities and events, as well as their interactions such as scene graphs and situation graphs, (ii) induction of event procedural knowledge to enhance vision-language pretraining through script learning, (iii) the distillation and encoding of the knowledge in language models, and (v) a wide range of multimedia understanding tasks that benefit from aforementioned techniques. We will conclude the tutorial with remaining challenges in this area. Basic knowledge of machine learning is required. An elementary understanding of pertaining techniques will be assumed. The tutorial will come along with a comprehensive survey, slides, videos, representative tools, data resources openly available at https://blender.cs.illinois.edu/tutorial/knowledgeVLP.

LSHP2: Automated Machine Learning & Tuning with FLAML
Chi Wang, Qingyun Wu, Xueqing, Luis Quintanilla

In this lab forum, we will provide an in-depth and hands-on tutorial on Automated Machine Learning & Tuning with a fast open-source library FLAML. The target audiences include data scientists, software engineers, and researchers. Attendees are assumed to have an introductory knowledge of either Python or .NET. Both ML beginners and AI experts can benefit from this tutorial. Attendees will learn how to use this lightweight library and related products to find accurate machine learning models automatically, efficiently, and economically. We will also introduce how to use FLAML to tune generic hyperparameters for a wide range of applications, such as MLOps workflows, pipelines, mathematical or statistical models, algorithm configurations, computing experiments, and software configurations. We will cover how to leverage the rich customization choices to meet your last-mile requirements for a successful deployment or apply this tool in your research. We will provide code examples, demos, and lessons learned from customers. The later parts of the tutorial will include advanced functionalities of the library, such as zero-shot AutoML, fair AutoML, and online AutoML. We will discuss open problems and challenges in the end.

This site is protected by copyright and trademark laws under US and International law. All rights reserved. Copyright © 1995–2023 AAAI