Recent work in synthetic data generation the time-series domain has focused on use of Generative Adversarial Networks. We propose a novel architecture for synthetically generating with Variational Auto-Encoders (VAEs). The proposed several distinct properties: interpretability, ability to encode knowledge, and reduced training times. evaluate quality by similarity predictability against four multivariate datasets. experiment varying sizes measure impact availability our VAE method as well...

While both the database and high-performance computing (HPC) communities utilize lossless compression methods to minimize floating-point data size, a disconnect persists between them. Each community designs assesses in domain-specific manner, making it unclear if HPC techniques can benefit applications or vice versa. With increasingly leaning towards in-situ analysis visualization, more from scientific simulations are being stored databases like Key-Value Stores queried using in-memory...

The existence of an anomaly detection method that is optimal for all domains a myth. Thus, there exists plethora methods which increases every year wide variety domains. But strength can also be weakness; given this massive library methods, how one select the best their application? Current literature focused on creating new or large frameworks experimenting with multiple at same time. However, and especially as continues to expand, extensive evaluation simply not feasible. To reduce burden,...

We investigate differences in user communication with live chat agents versus a commercial Intelligent Virtual Agent (IVA). This case study compares the two types of interactions same domain for company filling purposes. compared 16,794 human-to-human conversations and 27,674 IVA. Of those IVA conversations, 8,324 escalated to human agents. then investigated how strategies change when users first communicate an conversation thread. measured quantity, quality, diversity language, analyzed...

As Intelligent Virtual Agents (IVAs) increase in adoption and further emulate human personalities, we are interested how humans apply relational strategies to them compared other a service environment. Human-computer data from three live customer IVAs was collected, annotators marked all text that deemed unnecessary the determination of user intention as well presence multiple intents. After merging selections annotators, second round annotation determined classes language present sections...

We investigate the occurrence of user restatement when there is no apparent error in Intelligent Virtual Assistant (IVA) understanding a multimodal customer service environment. define several classes response medium combinations and use various statistical measures to determine how combination linguistic complexity impacts user's willingness accept their query result. Through analysis on 3; 000 sessions with live IVA deployed an airline company website mobile application, we discover that...

With the rise of Intelligent Virtual Assistants (IVAs), there is a necessary in human effort to identify conversations containing misunderstood user inputs. These uncover error natural language understanding and help prioritize expedite improvements IVA. As reviewer time valuable manual analysis consuming, prioritizing where misunderstanding has likely occurred reduces costs speeds improvement. In addition, less reviewed by humans mean data exposed, increasing privacy. We present scalable...

The dimensionality of traditional text representation is large, but the underlying data sparse. This makes clustering a very challenging task. Using language models and deep contextualized representations promising in many Natural Language Processing (NLP) tasks. However, some task-specific guidance necessary to adapt novel domain or particular downstream We present an empirical study pipeline for semi-supervised Our proposed method utilizes small number labeled samples fine-tune pre-trained...

We create and release the first publicly available commercial customer service corpus with annotated relational segments. Human-computer data from three live Intelligent Virtual Agents (IVAs) in domains of travel telecommunications were collected, reviewers marked all text that was deemed unnecessary to determination user intention. After merging selections multiple highlighted texts, a second round annotation done determine classes language present sections such as presence Greetings,...

The correct determination of user intent is key in dialog systems. However, an classifier often requires a large, labelled training dataset to identify set known intents. creation such complex and time-consuming task which usually involves humans applying clustering tools unlabelled data, analysing the results, creating human-readable labels for each cluster. While many Open Intent Discovery works tackle problem discovering clusters common intent, few generate label that can be used make...

We present a method to recognize states in remote classrooms provide autopilot services for distance education: no session, in-session, and question (i.e. student the classroom draws instructor's attention). study such that uses fuzzy classifiers above simple feature space presenting signal of human body movement. This computational model is largely inspired justified by previous studies on perception according Gestalt theory. Mass assignment theory (MAT) used constructing representing...

With the rise of intelligent virtual assistants (IVAs), there is a necessary in human effort to identify conversations containing misunderstood user inputs. These uncover error natural language understanding and help prioritize improvements IVA. As analysis time consuming expensive, prioritizing where misunderstanding has likely occurred reduces costs speeds IVA improvement. In addition, less reviewed by humans mean data are exposed, increasing privacy. We describe Trace AI, scalable system...

Abstract Research interest in Conversational artificial intelligence (ConvAI) has experienced a massive growth over the last few years and several recent advancements have enabled systems to produce rich varied turns conversations similar humans. However, this apparent creativity is also creating real challenge objective evaluation of such as authors are becoming reliant on crowd worker opinions primary measurement success and, so far, papers reporting all that necessary for others compare...

The existence of an anomaly detection method that is optimal for all domains a myth. Thus, there exists plethora methods which increases every year wide variety domains. But strength can also be weakness; given this massive library methods, how one select the best their application? Current literature focused on creating new or large frameworks experimenting with multiple at same time. However, and especially as continues to expand, extensive evaluation simply not feasible. To reduce burden,...

Abstract Claims have been made that speech recognition has achieved human parity, yet this does not appear to be the case in real‐world applications rely on it, especially for non‐native speakers. This then begs questions: What it even mean an AI system reach parity? How is progress towards goal being measured? article focuses current state of and recent developments benchmarking measuring performance models built processing. Through shift away from single metric benchmarks specialized...

Business managers using Intelligent Virtual Assistants (IVAs) to enhance their company's customer service need ways accurately and efficiently detect anomalies in conversations between the IVA customers, vital for retention satisfaction. Unfortunately, anomaly detection is a challenging problem because of subjective nature what defined as anomalous. Detecting sequences short texts, common chat settings, even more difficult independently generated texts are similar only at semantic level,...

The existence of a time series anomaly detection method that performs well for all domains is myth. Given massive library available methods, how can one select the best their application? An extensive evaluation every not feasible. Many existing systems do include an avenue human feedback, essential given subjective nature what even anomalous. We present technique improving univariate through automatic algorithm selection and human-in-the-loop false-positive removement. These determinations...