Topic models are widely explored for summarizing a corpus of documents. Recent advances in Variational AutoEncoder (VAE) have enabled the development black-box inference methods topic modeling order to alleviate drawbacks classical statistical inference. Most existing VAE based approaches assume unimodal Gaussian distribution approximate posterior latent variables, which limits flexibility encoding space. In addition, unsupervised architecture hinders incorporation extra label information,...

By jointly learning multiple tasks, multi-task (MTL) can leverage the shared knowledge across resulting in improved data efficiency and generalization performance. However, a major challenge MTL lies presence of conflicting gradients, which hinder fair optimization some tasks subsequently impede MTL's ability to achieve better overall Inspired by resource allocation communication networks, we formulate as utility maximization problem, where loss decreases are maximized under different...

Multi-task reinforcement learning (MTRL) has shown great promise in many real-world applications. Existing MTRL algorithms often aim to learn a policy that optimizes individual objective functions simultaneously with given prior preference (or weights) on different tasks. However, these methods suffer from the issue of \textit{gradient conflict} such tasks larger gradients dominate update direction, resulting performance degeneration other In this paper, we develop novel dynamic weighting...

Bilevel optimization has recently attracted considerable attention due to its abundant applications in machine learning problems. However, existing methods rely on prior knowledge of problem parameters determine stepsizes, resulting significant effort tuning stepsizes when these are unknown. In this paper, we propose two novel tuning-free algorithms, D-TFBO and S-TFBO. employs a double-loop structure with adaptively adjusted by the "inverse cumulative gradient norms" strategy. S-TFBO...

<div>Humans, as the most powerful learners on planet, have accumulated a lot of learning skills, such through tests, interleaving learning, self-explanation, active recalling, to name few. These skills and methodologies enable humans learn new topics more effectively efficiently. We are interested in investigating whether humans' can be borrowed help machines better. Specifically, we aim formalize these leverage them train better machine (ML) models. To achieve this goal, develop...

Interleaving learning is a human technique where learner interleaves the studies of multiple topics, which increases long-term retention and improves ability to transfer learned knowledge. Inspired by interleaving humans, in this paper we explore whether methodology beneficial for improving performance machine models as well. We propose novel framework referred (IL). In our framework, set collaboratively learn data encoder an fashion: trained model 1 while, then passed 2 further training, 3,...

The rapid development of e-commerce has led to the increasing role satisfaction in more fields. Therefore, customers' opinion become a necessary for success related companies. E-commerce satisfaction, as key factor affecting performance enterprises, research hotspot academia. This paper proposes synthetic evaluation model and logistics based on fuzzy dynamic weighted respectively. A modified ASCI analysis method structured equation is also proposed compare with method. Beyond this we have...

The rapid development of e-commerce has led to the increasing role satisfaction in more fields. Therefore, customers' opinion become a necessary for success related companies. E-commerce satisfaction, as key factor affecting performance enterprises, research hotspot academia. This paper proposes synthetic evaluation model and logistics based on fuzzy dynamic weighted respectively. A modified ASCI analysis method structured equation is also proposed compare with method. Beyond this we have...

<div>Humans, as the most powerful learners on planet, have accumulated a lot of learning skills, such through tests, interleaving learning, self-explanation, active recalling, to name few. These skills and methodologies enable humans learn new topics more effectively efficiently. We are interested in investigating whether humans' can be borrowed help machines better. Specifically, we aim formalize these leverage them train better machine (ML) models. To achieve this goal, develop...

Multi-objective optimization (MOO) has become an influential framework in many machine learning problems with multiple objectives such as criteria and multi-task (MTL). In this paper, we propose a new direction-oriented multi-objective problem by regularizing the common descent direction within neighborhood of that optimizes linear combination average loss MTL. This formulation includes GD MGDA special cases, enjoys benefit CAGrad, facilitates design stochastic algorithms. To solve problem,...