The upcoming talk covers a brief overview of fundamentals in optimal design construction and how metaheuristics can be used to design more efficient biomedical studies. Prof Wong Weng Kee will demonstrate the usefulness of nature-inspired metaheuristic algorithms for finding various hard-to-find optimal designs for complex models. They include locally optimal designs, minimax or standardised maximin optimal designs or Bayesian optimal designs.

As applications, Prof Wong will discuss search strategies to estimate some or all parameters in a compartment model, estimate the biological optimal dose in a continuation ratio model, construct multiple-objective optimal designs and extend Simon’s two-stage adaptive designs for Phase II trials to incorporate testing one of several pre-specified alternative hypotheses. 

The upcoming talk covers a brief overview of fundamentals in optimal design construction and how metaheuristics can be used to design more efficient biomedical studies. Prof Wong Weng Kee will demonstrate the usefulness of nature-inspired metaheuristic algorithms for finding various hard-to-find optimal designs for complex models. They include locally optimal designs, minimax or standardised maximin optimal designs or Bayesian optimal designs.

As applications, Prof Wong will discuss search strategies to estimate some or all parameters in a compartment model, estimate the biological optimal dose in a continuation ratio model, construct multiple-objective optimal designs and extend Simon’s two-stage adaptive designs for Phase II trials to incorporate testing one of several pre-specified alternative hypotheses. 

Intelligent vision tasks like visual recognition, object detection, tracking, and segmentation are being embedded into diverse intelligent systems including smartphones, self-driving cars, and smart factories due to the rapid development of deep learning techniques. Although reaching near-human-level performance, the state-of-the-art intelligent methods are developed on the pre-collected training and testing datasets. It is hard for them to get the same conclusions in the real world since complex and dynamic degradations cannot be covered by the datasets even if they are large-scale. Moreover, with the limited datasets, the robustness and security of the vision intelligent methods cannot be extensively evaluated and studied. 

Motivated by these challenges, Dr Guo Qing's research goal is to develop automatic techniques and practical tools for actively searching the vulnerability and evaluating the robustness of deep neural networks (DNNs) under diverse computer vision tasks and automatically enhancing or repairing the DNNs. Specifically, Dr Guo's group studies the vulnerability of examples from the perspective of adversarial attacks. The adversarial attack is an effective and fundamental way to reveal the vulnerability of DNNs. Nevertheless, most of the existing works focus on the additive-perturbation-based adversarial attack, neglecting other natural and important degradations, e.g., motion blur, light variation, weather variation, and vignetting, which widely exist in our daily life. In this talk, Dr Guo will share a series of natural-degradation-aware adversarial attacks including adversarial blur, rain, vignetting, exposure attacks, which cover a wide range of vision-based intelligent tasks, e.g., visual recognition, visual object tracking, object detection, etc. In addition, he will also explore benchmark construction for robustness evaluation. In terms of the DNN enhancement or repairing, Dr Guo will introduce their data restoration and augmentation techniques for enhancing image-based DNNs’ robustness, respectively.

K-Nearest Neighbors (kNN) is an important non-parametric statistical method, which is widely applied in many areas, such as functional estimation, density estimation and machine learning. Despite it’s popularity, some theoretical properties are still not well understood.

In this talk, Dr Zhao Puning will focus on three problems: 1) Estimation of entropy and mutual information; 2) Estimation of KL divergence; and 3) Minimax optimal adaptive classification and regression. For all of these problems, Dr Zhao will first derive the theoretical bound of the performance of the kNN method, and then derive the theoretical minimax lower bound of all methods using information theoretic approaches. Finally, he will talk about improvement on the traditional kNN method if the upper bound does not match the lower limit.

Towards the energy-efficient green AI goal, this talk focuses on the efficiency issues of the current dominant network architecture – Transformers. Although with excellent performance, the transformer is notorious for its costly training and inference due to its quadratic complexity of self-attention over a long sequence of representations, especially for high-resolution dense prediction. Prof Cai Jianfei will introduce a few simple yet effective solutions that can greatly improve the Transformer efficiency without causing a significant drop in performance.

Drug discovery is a very time-consuming and expensive process. For example, developing a new drug takes around ten years and could cost $2.5B. Recently, machine learning researchers have explored a way to accelerate such a process by automatically discovering (1) new molecules as drug candidates and (2) reaction pathways to synthesize the proposed molecule. 

In this talk, Prof Sungsoo Ahn will introduce the research progress of this area along with his recent works to tackle the main challenges.

In this talk, Prof Yeung Sai-Kit will be discussing his research relating to 3D reconstruction, scene understanding, content generation, and fabrication. He will also share this core research can be used in multi-disciplinary projects involving city planning, seafloor surveying, and fishery design. Lastly, he will conclude his talk by discussing potential collaborative projects between computer vision, graphics, and other disciplines to address challenging issues related to human empowerment and the building of sustainable environments. 

PID (Proportional, Integral and Derivative) Control is widely used in engineering systems, and has been viewed as a control method for linear systems with trail and error process for it’s gain determination. In this talk, Prof Song Yongduan will explore a performance-based method for systematically updating PID gains with guaranteed robustness and adaptability for nonlinear systems.

As exemplified by the COVID-19 pandemic, our health and well-being depend on a difficult-to-measure web of societal factors and individual behaviours. This effort requires new algorithmic and data-driven paradigms which span the full process of gathering costly data, learning models to understand and predict the interactions of many agents, and optimising the use of limited resources in interventions. 

In response to these needs, Dr Bryan Wilder will present methodological developments at the intersection of machine learning, optimisation, and social networks, which are motivated by on-the-ground collaborations on HIV prevention, tuberculosis treatment, and the COVID-19 response. 

Supervised learning requires the network to fit a fixed target function, hence value-based deep reinforcement learning (RL) agents must learn to predict a sequence of distinct value functions whose structure may change dramatically over the course of training. 

In this talk, Ms Clare Lyle will present a line of work which studies the implications of this observation on both the stability and the generalisation properties of value-based deep RL algorithms. 

In this talk, Dr Li Junnan will introduce several key directions to advance representation learning with few human annotations, namely (1) self-supervised learning, (2) semi-supervised learning, and (3) weakly-supervised vision-language pre-training. The talk will focus on state-of-the-art methods and recent developments.

In this talk, Dr Tay Yi, Co-founder & TechLead of the Unicorn AI team at Google Research, will introduce ExMix, an extreme mixture of 107 NLP tasks that pushes the boundaries of task scaling in NLP. 

Via extensive experiments, he will show that ExT5 outperforms strong T5 baselines on SuperGLUE, GEM, Rainbow, Closed-Book QA tasks, and several tasks outside of ExMix. ExT5 also significantly improves sample efficiency while pre-training.

In this talk, Mr David Berend from A*STAR explores how quantitative assessments can be used to study and enhance such trust by taking a close look at the robustness, privacy and security of deep neural networks. In particular, how the fundamental out-of-distribution technique can be helpful in providing insight for missing data, data generation and optimised neural architectures to achieve higher and trustworthy performance.

In this talk, Dr Zelda Mariet from Google Brain, will show that a simple dual reparameterisation of the bias-tradeoff drastically simplifies its analysis. 

Based on this reparameterisation, she will characterise the behaviour of bias and variance terms for a broad class of losses, aiming to provide a unified understanding of how different sources of randomness affect machine learning algorithms.

In this talk, Prof Masashi Sugiyama will first introduce the activities of the RIKEN Center for Advanced Intelligence Project (RIKEN-AIP). RIKEN is Japan’s largest and most comprehensive research organisation for basic and applied science, and AIP is working on advancing fundamental artificial intelligence technologies (machine learning, optimisation, etc.), their applications in accelerating scientific research (cancer, material, etc.), and solving socially critical problems (natural disaster, elderly healthcare, etc.), and discuss social aspects of artificial intelligence (ethical guidelines, personal data management, etc.).

In this talk, Mr. Kelvin Chan will first give a brief introduction to the recent developments of super-resolution. He will also present his efforts on effective image and video super-resolution and generalisation to real-world degradations through exploiting generative priors and temporal information.

In this installation, hear from Professor Sven Dickinson as he will be reviewing his efforts that draw on computer vision to understand the role that symmetry plays in human scene perception. Conversely, he will also look at how these results in human scene perception can strengthen the performance of modern deep learning computer vision systems for scene perception.