Aaron Li - Academia.edu (original) (raw)

Papers by Aaron Li

Styletts-VC: One-Shot Voice Conversion by Knowledge Transfer From Style-Based TTS Models

2022 IEEE Spoken Language Technology Workshop (SLT)

Discrete Applied Mathematics

Let D = d 1 , d 2 ,. .. , d n and F = f 1 , f 2 ,. .. , f n be two sequences of positive integers... more Let D = d 1 , d 2 ,. .. , d n and F = f 1 , f 2 ,. .. , f n be two sequences of positive integers. We consider the following decision problems: is there a i) multigraph, ii) loopless multigraph, iii) simple graph, iv) connected simple graph, v) tree, vi) caterpillar G = (V, E) such that for all k, d(v k) = d k and w∈N (v k) d(w) = f k (d(v) is the degree of v and N (v) is the set of neighbors of v). Here we show that all these decision problems can be solved in polynomial time if max k d k is bounded. The problem is motivated by NMR spectroscopy of hydrocarbons.

Engineering the Phenylpropanoid Pathway in Rhodosporidium toruloides for Naringenin Production from Tyrosine by Leveraging on its Native PAL Gene

arXiv (Cornell University), Oct 22, 2015

There is an explosion of data, documents, and other content, and people require tools to analyze ... more There is an explosion of data, documents, and other content, and people require tools to analyze and interpret these, tools to turn the content into information and knowledge. Topic modelling have been developed to solve these problems. Bayesian topic models such as Latent Dirichlet Allocation (LDA) [1] allow salient patterns in large collection of documents to be extracted and analyzed automatically. When analyzing texts, these patterns are called topics, represented as a distribution of words. Although numerous extensions of LDA have been created in academia in the last decade to address many problems, few of them can reliablily analyze multiple groups of documents and extract the similarities and differences in topics across these groups. Recently, the introduction of techniques for differential topic modelling, namely the Shadow Poisson Dirichlet Process model (SPDP) [2] performs uniformly better than many existing topic models in a discriminative setting. There is also a need to improve the running speed of algorithms for topic models. While some effort has been made for distributed algorithms, there is no work currently done using graphical processing units (GPU). Note the GPU framework has already become the most cost-efficient and popular parallel platform for many research and industry problems. In this thesis, I propose and implement a scalable multi-GPU distributed parallel framework which approximates SPDP, called MGPU-DP-SPDP, and a version running on a single GPU, Improved-GPU-SPDP. Through experiments, I have shown Improved-GPU-SPDP improved the running speed of SPDP by about 50 times while being almost as accurate as SPDP, with only one single cheap laptop GPU. Furthermore, I have shown the speed improvement of MGPU-DP-SPDP is sublinearly scalable when multiple GPUs are used, while keeping the accuracy fairly comparable to SPDP. Therefore, on a mediumsized GPU cluster, the speed improvement could potentially reach a factor of a thousand. Note SPDP is just a representative of perhaps another hundred other extensions of LDA. Although my algorithm is implemented to work with SPDP, it is designed to be a general framework that can be extended to work with other LDA extensions and improve

Follow Alice into the Rabbit Hole: Giving Dialogue Agents Understanding of Human Level Attributes

ArXiv, 2019

For conversational AI and virtual assistants to communicate with humans in a realistic way, they ... more For conversational AI and virtual assistants to communicate with humans in a realistic way, they must exhibit human characteristics such as expression of emotion and personality. Current attempts toward constructing human-like dialogue agents have presented significant difficulties. We propose Human Level Attributes (HLAs) based on tropes as the basis of a method for learning dialogue agents that can imitate the personalities of fictional characters. Tropes are characteristics of fictional personalities that are observed recurrently and determined by viewers’ impressions. By combining detailed HLA data with dialogue data for specific characters, we present a dataset that models character profiles and gives dialogue agents the ability to learn characters’ language styles through their HLAs. We then introduce a three-component system, ALOHA (which stands for Artificial Learning On Human Attributes), that combines character space mapping, character community detection, and language sty...

ArXiv, 2015

Latent variable models have accumulated a considerable amount of interest from the industry and a... more Latent variable models have accumulated a considerable amount of interest from the industry and academia for their versatility in a wide range of applications. A large amount of effort has been made to develop systems that is able to extend the systems to a large scale, in the hope to make use of them on industry scale data. In this paper, we describe a system that operates at a scale orders of magnitude higher than previous works, and an order of magnitude faster than state-of-the-art system at the same scale, at the same time showing more robustness and more accurate results. Our system uses a number of advances in distributed inference: high performance in synchronization of sufficient statistics with relaxed consistency model; fast sampling, using the Metropolis-Hastings-Walker method to overcome dense generative models; statistical modeling, moving beyond Latent Dirichlet Allocation (LDA) to Pitman-Yor distributions (PDP) and Hierarchical Dirichlet Process (HDP) models; sophist...

Journal of Nursing Education and Practice, 2016

Undergraduate nursing curriculum is changing to keep pace with the healthcare system. As a result... more Undergraduate nursing curriculum is changing to keep pace with the healthcare system. As a result, nursing faculties must consider innovative approaches to clinical instruction. In 2010, one nursing faculty transformed the traditional sessional clinical instructor role into a Nursing Practice Instructor role in order to facilitate the integration between theory and practice in both on and off campus settings. This descriptive qualitative study involved conversational interviews led by Nursing Practice Instructor peer-researchers to elicit the perceptions of how roles have changed from that of a sessional instructor. Eligibility for participation included all Nursing Practice Instructors who previously held a role as a sessional instructor in the same faculty. Data Analysis was done using a content analysis approach where themes within each guiding question were identified and then compared for congruency and further interpretation. Participants felt that there were differences between the sessional Clinical Instructors and Nursing Practice Instructor roles and expectations and as a result of this change, they were more invested in their teaching role based on their ability to integrate the curriculum, the opportunity to engage in the faculty, and contribute to student learning in a more significant way. Overall, the Nursing Practice Instructor role has initiated changes in how clinical instructors are employed and supported, contributing positively to the outcomes associated with an integrated, context-relevant curriculum, and ultimately, fostering future nurses with the ability to make a difference in the healthcare system.

2015 Australasian Universities Power Engineering Conference (AUPEC), 2015

The mismatch between electricity supply and demand in power system and the management of it using... more The mismatch between electricity supply and demand in power system and the management of it using response loads leads to the notion of Demand Response. In this paper, a comprehensive review of Multi-Agent System is carried out along with works in the space of demand response and power system operation. Factoring from different scopes, two aspects are thereafter highlighted-MAS-based DR in market systems and distribution network assets, respectively. Alongside, a comparison of popular tools used for demand response and impact analysis is also addressed since proper tools can be more beneficial and recognizable to effective implementation. The benefits of the various actors towards MAS technology in the transition towards Smart Grid are also highlighted. Finally, some unsolved questions and remaining challenges are explictly identified.