In this paper, we explore the slot tagging with only a few labeled support sentences (a.k.a. few-shot). Few-shot faces unique challenge compared to other fewshot classification problems as it calls for modeling dependencies between labels. But is hard apply previously learned label an unseen domain, due discrepancy of sets. To tackle this, introduce collapsed dependency transfer mechanism into conditional random field (CRF) abstract patterns transition scores. few-shot setting, emission...

Real-world data usually couples the label ambiguity and heavy imbalance, challenging algorithmic robustness of partial learning (PLL) long-tailed (LT). The straightforward combination LT PLL, i.e., LT-PLL, suffers from a fundamental dilemma: methods build upon given class distribution that is unavailable in performance PLL severely influenced context. We show even with auxiliary an oracle prior, state-of-the-art underperform due to adverse fact constant rebalancing harsh disambiguation PLL....

In this paper, we introduce an event-driven trading strategy that predicts stock movements by detecting corporate events from news articles.Unlike existing models utilize textual features (e.g., bag-of-words) and sentiments to directly make predictions, consider as the driving force behind aim profit temporary mispricing may occur when take place.The core of proposed is a bi-level event detection model.The low-level detector identifies events' existences each token, while high-level...

With the rapid development of deep learning in recent years, recommendation algorithm combined with model has become an important direction field future. Personalized resource is main way to realize students’ adaptation system. Based on in-depth mode, online action data are obtained, and further analysis technology used construct special mode provide suitable resources. The traditional method introducing resources mainly stays at level examination questions. What ignores essence knowledge...

Zhihan Zhou, Dejiao Zhang, Wei Xiao, Nicholas Dingwall, Xiaofei Ma, Andrew Arnold, Bing Xiang. Proceedings of the 2022 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2022.

Learning with noisy labels has become imperative in the Big Data era, which saves expensive human labors on accurate annotations. Previous noise-transition-based methods have achieved theoretically-grounded performance under Class-Conditional Noise model (CCN). However, these approaches builds upon an ideal but impractical anchor set available to pre-estimate noise transition. Even though subsequent works adapt estimation as a neural layer, ill-posed stochastic learning of its parameters...

ABSTRACT Background Computational drug repurposing is a cost- and time-efficient approach that aims to identify new therapeutic targets or diseases (indications) of existing drugs/compounds. It especially critical for emerging and/or orphan due its cheaper investment shorter research cycle compared with traditional wet-lab discovery approaches. However, the underlying mechanisms action (MOAs) between repurposed drugs their target remain largely unknown, which still main obstacle...

While few-shot classification has been widely explored with similarity based methods, sequence labeling poses a unique challenge as it also calls for modeling the label dependencies. To consider both item and dependency, we propose to leverage conditional random fields (CRFs) in labeling. It calculates emission score methods obtains transition specially designed transfer mechanism. When applying CRF scenarios, discrepancy of sets among different domains makes hard use dependency learned...

In this paper, we explore the slot tagging with only a few labeled support sentences (a.k.a. few-shot). Few-shot faces unique challenge compared to other few-shot classification problems as it calls for modeling dependencies between labels. But is hard apply previously learned label an unseen domain, due discrepancy of sets. To tackle this, introduce collapsed dependency transfer mechanism into conditional random field (CRF) abstract patterns transition scores. setting, emission score CRF...

Speaker diarization (detecting who-spoke-when using relative identity labels) and speaker recognition absolute labels without timing) are different but related tasks that often need to be completed simultaneously in many scenarios. Traditional methods, however, address them independently. In this paper, we propose a method jointly diarize recognize speakers from collection of conversations. This benefits the sparsity temporal smoothness within conversation large-scale timbre modeling across...

In order to analyze the supervision mode of platform economy from perspective blockchain, a blockchain-based method is proposed for economy. First, in terms speed data collection, economic monitoring and forecasting based on network big are real-time acquisition analysis, so that problems can be detected earlier serve purpose early warning forecasting; secondly, scope monitoring, has not been monitored traditional fields becomes possible; thirdly, mining it reflect more unpredictable...

User embeddings play a crucial role in user engagement forecasting and personalized services. Recent advances sequence modeling have sparked interest learning from behavioral data. Yet behavior-based embedding faces the unique challenge of dynamic modeling. As users continuously interact with apps, should be periodically updated to account for users' recent long-term behavior patterns. Existing methods highly rely on stateless models that lack memory historical behavior. They either discard...

Robust reinforcement learning (RL) aims to find a policy that optimizes the worst-case performance in face of uncertainties. In this paper, we focus on action robust RL with probabilistic execution uncertainty, which, instead always carrying out specified by policy, agent will take probability $1-\rho$ and an alternative adversarial $\rho$. We establish existence optimal MDPs uncertainty provide Bellman optimality equation for its solution. Furthermore, develop Action Reinforcement Learning...

Self-supervised learning (SSL) as an effective paradigm of representation has achieved tremendous success on various curated datasets in diverse scenarios. Nevertheless, when facing the long-tailed distribution real-world applications, it is still hard for existing methods to capture transferable and robust representation. Conventional SSL methods, pursuing sample-level uniformity, easily leads disparity where head classes dominate feature regime but tail passively collapse. To address this...

In this paper, we introduce an event-driven trading strategy that predicts stock movements by detecting corporate events from news articles. Unlike existing models utilize textual features (e.g., bag-of-words) and sentiments to directly make predictions, consider as the driving force behind aim profit temporary mispricing may occur when take place. The core of proposed is a bi-level event detection model. low-level detector identifies events' existences each token, while high-level...

Self-supervised learning has achieved a great success in the representation of visual and textual data. However, current methods are mainly validated on well-curated datasets, which do not exhibit real-world long-tailed distribution. Recent attempts to consider self-supervised made by rebalancing loss perspective or model perspective, resembling paradigms supervised learning. Nevertheless, without aid labels, these explorations have shown expected significant promise due limitation tail...

Learning with noisy labels has become imperative in the Big Data era, which saves expensive human labors on accurate annotations. Previous noise-transition-based methods have achieved theoretically-grounded performance under Class-Conditional Noise model (CCN). However, these approaches builds upon an ideal but impractical anchor set available to pre-estimate noise transition. Even though subsequent works adapt estimation as a neural layer, ill-posed stochastic learning of its parameters...

In human-computer interaction, understanding user behaviors and tailoring systems accordingly is pivotal. To this end, general-purpose representation learning based on behavior logs emerging as a powerful tool in modeling, offering adaptability to various downstream tasks such item recommendations ad conversion prediction, without the need fine-tune upstream model. While methodology has shown promise contexts like search engines e-commerce platforms, its fit for instant messaging apps,...