Chenliang Zhou

Chenliang Zhou

PhD student at University of Cambridge

Posts by Collection

portfolio

publications

AISpace2: An Interactively Visualizable Tool for Learning and Teaching Artificial Intelligence Permalink

Published in AAAI, 2020

AIspace is a set of tools used to learn and teach fundamental AI algorithms. The original version of AIspace was written in Java, and there was not a clean separation of the algorithms and visualization; it was too complicated for students to modify the underlying algorithms. Its next generation, AIspace2, is built on AIPython, an open source Python code that is designed to be as close as possible to pseudocode. AISpace2, visualized in JupyterLab, keeps the simple Python code, and uses the hooks in AIPython to allow visualization of the algorithms. This allows students to see and modify the high-level algorithms in Python, and to visualize the output in a graphical form, hence better helps them to build confidence and comfort in AI concepts and algorithms. So far we have tools for search, constraint satisfaction problems (CSP), planning and Bayesian network. In this paper we outline the tools and give some evaluations based on user feedback.

Investigation on Circadian Action and Color Quality in Laser-Based Illuminant for General Lighting and Display Permalink

Published in IEEE Photonics Journal, 2020

In this work, the genetic algorithm (GA) is employed to optimize both circadian action factor (CAF) and color quality of laser-based illuminants (LBIs) with three, four, and ve spectral bands to disclose its possible use in two common white lighting applications, i.e. bedroom lighting and ofce lighting. Comparing all LBIs at a correlated color temperature (CCT) of 3000 K and a color rendering index of 80, the CAF of four-band LBIs reaches a minimum of 0.238 and maintains at a possibly highest luminous efcacy of radiation (LER) of 422 lm/W among all cases. The performances of white LBIs are also compared with those of white light-emitting diodes (LEDs). The results demonstrate that, under the same conditions of color rendering and color temperature, both four-band LBIs and fourband LEDs exhibit the largest circadian tunability of about 4.7, while four-band LBIs possess much higher LER at the same time compared with four-band LEDs. In addition, for the display application, the investigation on the optimal circadian tunability as a function of color gamut at two CCTs (3000 K and 6500 K) is also performed.

XQSV: A Structurally Variable Network to Imitate Human Play in Xiangqi

Published in IEEE CoG, 2024

In this paper, we introduce an innovative deep learning architecture, termed Xiangqi Structurally Variable (XQSV), designed to emulate the behavioral patterns of human players in Xiangqi, or Chinese Chess. The unique attribute of XQSV is its capacity to alter its structural configuration dynamically, optimizing performance for the task based on the particular subset of data on which it is trained. We have incorporated several design improvements to significantly enhance the network’s predictive accuracy, including a local illegal move filter, an Elo range partitioning, a sequential one-dimensional input, and a simulation of imperfect memory capacity. Empirical evaluations reveal that XQSV attains a predictive accuracy of approximately 40%, with its performance peaking within the trained Elo range. This indicates the model’s success in mimicking the play behavior of individuals within that specific range. A three-terminal Turing Test was employed to demonstrate that the XQSV model imitates human behavior more accurately than conventional Xiangqi engines, rendering it indistinguishable from actual human opponents. Given the inherent nondeterminism in human gameplay, we propose two supplementary relaxed evaluation metrics. To our knowledge, XQSV represents the first model to mimic Xiangqi players.

Fitness Aware Human Motion Generation from Fine-Tuning

Published in NeurIPS, 2024

Diffusion models have recently gained considerable attention in 3D human motion generation due to their ability to handle complex human movements. However, existing models fail to incorporate the nuances presented by individual physical fitness levels. Therefore, we address this gap by integrating Functional Movement Screen (FMS) scores into diffusion models through fine-tuning, enabling the generation of fitness-aware motions. This approach transforms FMS data into HumanML3D format, optimises a base diffusion model, and introduces conditioning based on FMS scores. As a result, our fine-tuned model is capable of generating motions tailored to individual fitness levels and shows significant improvements in motion generation fidelity. Producing synthetic human motions conditioned on fitness levels is a novel approach that can be highly beneficial for various fields such as healthcare, sports, and entertainment.

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.