arXiv subjects

Xiaomin Li

Publications and source records attributed to Xiaomin Li.

59 records · Page 4Linked to original sources

arXiv2022

TTS-GAN: A Transformer-based Time-Series Generative Adversarial Network

Signal measurements appearing in the form of time series are one of the most common types of data used in medical machine learning applications. However, such datasets are often small, making the training of deep neural network architectures ineffective. For time-series, the suite of data augmentation tricks we can use to expand the size of the dataset is limited by the need to maintain the basic properties of the signal. Data generated by a Generative Adversarial Network (GAN) can be utilized as another data augmentation tool. RNN-based GANs suffer from the fact that they cannot effectively model long sequences of data points with irregular temporal relations. To tackle these problems, we introduce TTS-GAN, a transformer-based GAN which can successfully generate realistic synthetic time-series data sequences of arbitrary length, similar to the real ones. Both the generator and discriminator networks of the GAN model are built using a pure transformer encoder architecture. We use visualizations and dimensionality reduction techniques to demonstrate the similarity of real and generated time-series data. We also compare the quality of our generated data with the best existing alternative, which is an RNN-based time-series GAN.

cs.LG↗

arXiv2021

Webster Sequences, Apportionment Problems, and Just-In-Time Sequencing

Given a real number $α\in (0,1)$, we define the Webster sequence of density $α$ to be $W_α= (\lceil(n-1/2) / α\rceil)_{n\in\mathbb{N}}$, where $\lceil x \rceil$ is the ceiling function. It is known that if $α$ and $β$ are irrational with $α+ β= 1$, then $W_α$ and $W_β$ partition $\mathbb{N}$. On the other hand, an analogous result for three-part partitions does not hold: There does not exist a partition of $\mathbb{N}$ into sequences $W_α, W_β, W_γ$ with $α, β, γ$ irrational. In this paper, we consider the question of how close one can come to a three-part partition of $\mathbb{N}$ into Webster sequences with prescribed densities $α, β, γ$. We show that if $α, β, γ$ are irrational with $α+ β+γ= 1$, there exists a partition of $\mathbb{N}$ into sequences of densities $α, β, γ$, in which one of the sequences is a Webster sequence and the other two are "almost" Webster sequences that are obtained from Webster sequences by perturbing some elements by at most 1. We also provide two efficient algorithms to construct such partitions. The first algorithm outputs the $n$th element of each sequence in $O(1)$ time and the second algorithm gives the assignment of the $m$th positive integer to a sequence in $O(1)$ time. We show that the results are best-possible in several respects. Moreover, we describe applications of these results to apportionment and optimal scheduling problems.

math.NT↗

arXiv2021

Manipulation of polar vortex chirality in oxide superlattices

Topological polar vortices that are the electric analogues of magnetic objects, present great potential in applications of future nanoelectronics due to their nanometer size, anomalous dielectric response, and chirality. To enable the functionalities, it is prerequisite to manipulate the polar states and chirality by using external stimuli. Here, we probe the evolutions of polar state and chirality of polar vortices in PbTiO3/SrTiO3 superlattices under electric field by using atomically resolved in situ scanning transmission electron microscopy and phase-field simulations. We find that the adjacent clockwise and counterclockwise vortex usually have opposite chirality. The phase-field simulations suggest that the rotation reversal or axial polarization switching can lead to the chirality change. Guided by which, we experimentally validate that the vortex rotation direction can be changed by applying and subsequently removing of electric fields, offering a potential strategy to manipulate the vortex chirality. The revealed details of dynamic behavior for individual polar vortices at atomic scale and the proposed strategy for chirality manipulation provide fundamentals for future device applications.

cond-mat.mtrl-sci↗

arXiv2020

Parallel Blockwise Knowledge Distillation for Deep Neural Network Compression

Deep neural networks (DNNs) have been extremely successful in solving many challenging AI tasks in natural language processing, speech recognition, and computer vision nowadays. However, DNNs are typically computation intensive, memory demanding, and power hungry, which significantly limits their usage on platforms with constrained resources. Therefore, a variety of compression techniques (e.g. quantization, pruning, and knowledge distillation) have been proposed to reduce the size and power consumption of DNNs. Blockwise knowledge distillation is one of the compression techniques that can effectively reduce the size of a highly complex DNN. However, it is not widely adopted due to its long training time. In this paper, we propose a novel parallel blockwise distillation algorithm to accelerate the distillation process of sophisticated DNNs. Our algorithm leverages local information to conduct independent blockwise distillation, utilizes depthwise separable layers as the efficient replacement block architecture, and properly addresses limiting factors (e.g. dependency, synchronization, and load balancing) that affect parallelism. The experimental results running on an AMD server with four Geforce RTX 2080Ti GPUs show that our algorithm can achieve 3x speedup plus 19% energy savings on VGG distillation, and 3.5x speedup plus 29% energy savings on ResNet distillation, both with negligible accuracy loss. The speedup of ResNet distillation can be further improved to 3.87 when using four RTX6000 GPUs in a distributed cluster.

cs.LG↗

arXiv2019

Almost Beatty Partitions

Given $0<α<1$, the Beatty sequence of density $α$ is the sequence $B_α=(\lfloor n/α\rfloor)_{n\in\mathbb{N}}$. Beatty's theorem states that if $α,β$ are irrational numbers with $α+β=1$, then the Beatty sequences $B_α$ and $B_β$ partition the positive integers, that is, each positive integer belongs to exactly one of these two sequences. On the other hand, Uspensky showed that this result breaks down completely for partitions into three (or more) sequences: There does not exist a single triple $(α,β,γ)$ such that the Beatty sequences $B_α,B_β,B_γ$ partition the positive integers. In this paper we consider the question of how close we can come to a three-part Beatty partition by considering "almost" Beatty sequences, that is, sequences that represent small perturbations of an "exact" Beatty sequence. We first characterize all cases in which there exists a partition into two exact Beatty sequences and one almost Beatty sequence with given densities, and we determine the approximation error involved. We then give two general constructions that yield partitions into one exact Beatty sequence and two almost Beatty sequences with prescribed densities, and we determine the approximation error in these constructions. Finally, we show that in many situations these constructions are best-possible in the sense that they yield the closest approximation to a three-part Beatty partition.

math.NT↗