MultiWOZ is a recently-released multidomain dialogue dataset spanning 7 distinct domains and containing over 10000 dialogues, one of the largest resources of its kind to-date. Though an immensely useful resource, while building different classes of dialogue state tracking models using MultiWOZ, we detected substantial errors in the state annotations and dialogue utterances which negatively impacted the performance of our models. In order to alleviate this problem, we use crowdsourced workers to fix the state annotations and utterances in the original version of the data. Our correction process results in changes to over 32% of state annotations across 40% of the dialogue turns. In addition, we fix 146 dialogue utterances throughout the dataset focusing in particular on addressing slot value errors represented within the conversations. We then benchmark a number of state-of-the-art dialogue state tracking models on this new Mul-tiWOZ 2.1 dataset and show joint state tracking performance on the corrected state annotations. We are publicly releasing MultiWOZ 2.1 to the community, hoping that this dataset resource will allow for more effective dialogue state tracking models to be built in the future.

Dialog state tracking is used to estimate the current belief state of a dialog given all the preceding conversation. Machine reading comprehension, on the other hand, focuses on building systems that read passages of text and answer questions that require some understanding of passages. We formulate dialog state tracking as a reading comprehension task to answer the question what is the state of the current dialog? after reading conversational context. In contrast to traditional state tracking methods where the dialog state is often predicted as a distribution over a closed set of all the possible slot values within an ontology, our method uses a simple attention-based neural network to point to the slot values within the conversation. Experiments on MultiWOZ-2.0 cross-domain dialog dataset show that our simple system can obtain similar accuracies compared to the previous more complex methods. By exploiting recent advances in contextual word embeddings, adding a model that explicitly tracks whether a slot value should be carried over to the next turn, and combining our method with a traditional joint state tracking method that relies on closed set vocabulary, we can obtain a joint-goal accuracy of 47.33% on the standard test split, exceeding current state-of-the-art by 11.75%**.

Purpose To perform a comprehensive validation of plans generated on a preconfigured Halcyon 2.0 with preloaded beam model, including evaluations of new features and implementing the patient specific quality assurance (PSQA) process with multiple detectors. Methods A total of 56 plans were generated in Eclipse V15.6 (Varian Medical System) with a preconfigured Halcyon treatment machine. Ten plans were developed via the AAPM TG‐119 test suite with both IMRT and VMAT techniques. 34 clinically treated plans using C‐arm LINAC from 24 patients were replanned on Halcyon using IMRT or VMAT techniques for a variety of sites including: brain, head and neck, lung, breast, abdomen, and pelvis. Six of those plans were breast VMAT plans utilizing the extended treatment field technique available with Halcyon 2.0. The dynamically flattened beam (DFB), another new feature on Halcyon 2.0, was also used for an AP/PA spine and four field box pelvis, as well as ten 3D breast plans. All 56 plans were measured with an ion chamber (IC), film, portal dosimetry (PD), ArcCHECK, and Delta4. Tolerance and action limits were calculated and compared to the recommendations of TG‐218. Results TG‐119 IC and film confidence limits met those set by the task group, except for IMRT target point dose. Forty‐four of 46 clinical plans were within 3% for IC measurements. Average gamma passing rates with 3% dose difference and 2mm distance‐to‐agreement for IMRT/VMAT plans were: Film – 96.8%, PD – 99.9%, ArcCHECK – 99.1%, and Delta4 – 99.2%. Calculated action limits were: Film – 86.3%, PD – 98.4%, ArcCHECK – 96.1%, and Delta4 – 95.7%. Extended treatment field technique was fully validated and 3D plans with DFB had similar results to IMRT/VMAT plans. Conclusion Halcyon plan deliveries were verified with multiple measurement devices. New features of Halcyon 2.0 were also validated. Traditional PSQA techniques and process specific tolerance and action limits were successfully implemented.

Location services, fundamentally, rely on two components: a mapping system and a positioning system. The mapping system provides the physical map of the space, and the positioning system identifies the position within the map. Outdoor location services have thrived over the last couple of decades because of well-established platforms for both these components (e.g. Google Maps for mapping, and GPS for positioning). In contrast, indoor location services haven't caught up because of the lack of reliable mapping and positioning frameworks. Wi-Fi positioning lacks maps and is also prone to environmental errors. In this paper, we present DLoc, a Deep Learning based wireless localization algorithm that can overcome traditional limitations of RF-based localization approaches (like multipath, occlusions, etc.). We augment DLoc with an automated mapping platform, MapFind. MapFind constructs location-tagged maps of the environment and generates training data for DLoc. Together, they allow off-the-shelf Wi-Fi devices like smartphones to access a map of the environment and to estimate their position with respect to that map. During our evaluation, MapFind has collected location estimates of over 105 thousand points under 8 different scenarios with varying furniture positions and people motion across two different spaces covering 2000 sq. Ft. DLoc outperforms stateof-the-art methods in Wi-Fi-based localization by 80% (median & 90 th percentile) across the two different spaces. CCS CONCEPTS • Networks → Location based services; • Computing methodologies → Robotic planning; Supervised learning; • Information systems → Sensor networks.

Data availability is a bottleneck during early stages of development of new capabilities for intelligent artificial agents. We investigate the use of text generation techniques to augment the training data of a popular commercial artificial agent across categories of functionality, with the goal of faster development of new functionality. We explore a variety of encoderdecoder generative models for synthetic training data generation and propose using conditional variational auto-encoders. Our approach requires only direct optimization, works well with limited data and significantly outperforms the previous controlled text generation techniques. Further, the generated data are used as additional training samples in an extrinsic intent classification task, leading to improved performance by up to 5% absolute f-score in low-resource cases, validating the usefulness of our approach.