In this paper, we compare two approaches to train a multilingual language model: (i) simple multilingual learning using data-mixing, and (ii) meta-learning. We examine the performance of these models by extending them to unseen language pairs and further finetune them for the task of unsupervised NMT. We perform several experiments with varying amounts of data and give a comparative analysis of the approaches. We observe that both approaches give a comparable performance, and meta-learning gives slightly better results in a few cases of low amounts of data. For Oriya-Punjabi language pair, meta-learning performs better than multilingual learning when using 2M, and 3M sentences.

Site Reliability Engineers (SREs) play a key role in identifying the cause of an issue and preforming remediation steps to resolve it. After an issue is reported, SREs come together in a virtual room (collaboration platform) to triage the issue. While doing so, they leave behind a wealth of information, in the form of conversations, which can be used later for triaging similar issues. However, usability of these conversations offer challenges due to them being and scarcity of conversation utterance label. This paper presents a novel approach for issue artefact extraction from noisy conversations with minimal labelled data. We propose a combination of unsupervised and supervised models with minimal human intervention that leverages domain knowledge to predict artefacts for a small amount of conversation data and use that for fine-tuning an already pre-trained language model for artefact prediction on a large amount of conversation data. Experimental results on our dataset show that the proposed ensemble of the unsupervised and supervised models is better than using either one of them individually. We also present a deployment case study of the proposed artefact prediction.

The noisiness and low quality of IT operations management data is a major challenge in using machine learning to assist IT operations management. Our system mitigates this challenge by automatically measuring data quality, and then using the results to select data subsets that generate improved model performance. Based on a set of metrics that quantify the quality of a corpus with both structured and unstructured data, we are proposing a framework to automatically identify "well behaved" subsets in the corpus. By streaming input data to separate models for these subsets, we can achieve better performance when compared with a model trained on the full dataset. We present a motivating example that inspired our approach as well as a deployment case study of our system based on engagements with two clients which demonstrate that the proposed methodology is effective for detecting such subsets to improve model performance.

Contrastive explanations for understanding the behavior of black box models has gained a lot of attention recently as they provide potential for recourse.In this paper, we propose a method Contrastive Attributed explanations for Text (CAT) which provides contrastive explanations for natural language text data with a novel twist as we build and exploit attribute classifiers leading to more semantically meaningful explanations.To ensure that our contrastive generated text has the fewest possible edits with respect to the original text, while also being fluent and close to a human generated contrastive, we resort to a minimal perturbation approach regularized using a BERT language model and attribute classifiers trained on available attributes.We show through qualitative examples and a user study that our method not only conveys more insight because of these attributes, but also leads to better quality (contrastive) text.Quantitatively, we show that our method outperforms other state-ofthe-art methods across four data sets on four benchmark metrics.

Change is one of the biggest contributors to service outages. With more enterprises migrating their applications to cloud and using automated build and deployment the volume and rate of changes has significantly increased. Furthermore, microservice-based architectures have reduced the turnaround time for changes and increased the dependency between services. All of the above make it impossible for the Site Reliability Engineers (SREs) to use the traditional methods of manual risk assessment for changes. In order to mitigate change-induced service failures and ensure continuous improvement for cloud native services, it is critical to have an automated system for assessing the risk of change deployments. In this paper, we present an AI-based system for proactively assessing the risk associated with deployment of application changes in cloud operations. The risk assessment is accompanied with actionable risk explainability. We discuss the usage of this system in two primary scenarios of automated and manual deployment. In automated deployment scenario, our approach is able to alert SREs on 70 % of problematic changes by blocking only 1.5 % of total changes and recommending human intervention. In manual deployment scenario, our approach recommends the SREs to perform extra due diligence for 2.8 % of total changes to capture 84 % of problematic changes.

Conversational channels are changing the landscape of hybrid cloud service management. These channels are becoming important avenues for Site Reliability Engineers (SREs) %Subject Matter Experts (SME) to collaboratively work together to resolve an incident or issue. Identifying segmented conversations and extracting key insights or artefacts from them can help engineers to improve the efficiency of the incident remediation process by using information retrieval mechanisms for similar incidents. However, it has been empirically observed that due to the semi-formal behavior of such conversations (human language) the conversations are very unique in nature and also contain domain-specific terms. %It is important to identify the correct keywords and artefacts like symptoms, issue etc., present in the conversation chats. In this paper, we build a framework that taps into the conversational channels and uses various learning methods to (1) understand and extract key artefacts from conversations like diagnostic steps and resolution actions taken and (2) present an approach to identify past conversations about similar issues. Experimental results on our dataset show the efficacy of the methods used in our proposed system.

We propose a novel framework for controllable natural language transformation. Realizing that the requirement of parallel corpus is practically unsustainable for controllable generation tasks, an unsupervised training scheme is introduced. The crux of the framework is a deep neural encoder-decoder that is reinforced with text-transformation knowledge through auxiliary modules (called scorers). These scorers, based on off-the-shelf language processing tools, decide the learning scheme of the encoder-decoder based on its actions. We apply this framework for the text-transformation task of formalizing an input text by improving its readability grade; the degree of required formalization can be controlled by the user at run-time. Experiments on public datasets demonstrate the efficacy of our model towards: (a) transforming a given text to a more formal style, and (b) varying the amount of formalness in the output text based on the specified input control. Our code and datasets are released for academic use.

The impact of soiling on solar panels is an important and well-studied problem in renewable energy sector. In this paper, we present the first convolutional neural network (CNN) based approach for solar panel soiling and defect analysis. Our approach takes an RGB image of solar panel and environmental factors as inputs to predict power loss, soiling localization, and soiling type. In computer vision, localization is a complex task which typically requires manually labeled training data such as bounding boxes or segmentation masks. Our proposed approach consists of specialized four stages which completely avoids localization ground truth and only needs panel images with power loss labels for training. The region of impact area obtained from the predicted localization masks are classified into soiling types using the webly supervised learning. For improving localization capabilities of CNNs, we introduce a novel bi-directional input-aware fusion (BiDIAF) block that reinforces the input at different levels of CNN to learn input-specific feature maps. Our empirical study shows that BiDIAF improves the power loss prediction accuracy by about 3% and localization accuracy by about 4%. Our end-to-end model yields further improvement of about 24% on localization when learned in a weakly supervised manner. Our approach is generalizable and showed promising results on web crawled solar panel images. Our system has a frame rate of 22 fps (including all steps) on a NVIDIA TitanX GPU. Additionally, we collected first of it's kind dataset for solar panel image analysis consisting 45,000+ images.

One of the major challenges confronting the widespread adoption of solar energy is the uncertainty of production. The energy generated by photo-voltaic systems is a function of the received solar irradiance which varies due to atmospheric and weather conditions. A key component required for forecasting irradiance accurately is the clear sky model which estimates the average irradiance at a location at a given time in the absence of clouds. Current methods for modelling clear sky irradiance are either inaccurate or require extensive atmospheric data, which tends to vary with location and is often unavailable. In this paper, we present a data-driven methodology, Blue Skies, for modelling clear sky irradiance solely based on historical irradiance measurements. Using machine learning techniques, Blue Skies is able to generate clear sky models that are more accurate spatio-temporally compared to the state of the art, reducing errors by almost 50%.

We study an offloading mechanism for cellular networks in which mobiles with good cellular links can act as hotspots and can assist other mobiles. We study throughput optimal offloading and also a fair offloading strategy which we call proportional increment offloading. We show that the former problem can be reduced to a capacitated facility location problem (CFLP) whereas the latter can be solved by solving a sequence of CFLPs. We propose a belief propagation based algorithm to solve CFLP, a well known NP complete problem. We primarily consider point coordination function (PCF) based WiFi access for offloading but also discuss distributed coordination function (DCF) based access and related issues. Further, we argue that all the mobiles benefit through participating in offloading. We perform extensive simulation to evaluate the performance of the proposed algorithms and effectiveness of mobile assisted offloading.

MQTT-SN offers energy efficient data transportation under unreliable wireless environment with limited energy/ bandwidth. However, MQTT-SN faces a new set of challenges for service guarantees when operated for a reliable or time-critical applications. The impact of various network parameters on the MQTT-SN service assurance requires an end-to-end system study, which are unexplored yet. In this paper we analyze in detail about the end-to-end service assurance parameters such as content delivery delay and probability of content delivery for a MQTT-SN to be used in health care Internet of Things (IoT). To model the end-to-end delay, various entities of the MQTT-SN including handshake messages in MQTT-SN Pub/sub architecture over TCP, over UDP and then the queuing delay involved in server have been considered. We model the MQTT content and request server as a round-robin queue scheduler and also model a TOPIC ID based matching station that delivers the content on requested TOPIC ID. With this model we derive the end-to-end content delivery probability and content delivery delay estimate as functions of MQTT-SN system parameters such as content arrival rate, request arrival rate, possession time of content in server, possession time of request in server, number of content arrivals and number of requests arrivals. Both these service assurance parameters will give more insight into the number of content/request arrivals that can be supported for a given service assurance requirement. In addition these service assurance parameters will be of help to design the system more effectively. We carry out a detailed ns2 based simulation to show impacts of various system parameters on the network performances for various QoS services designed for MQTT-SN system.

Non-beacon mode operation is included in IEEE 802.1.5.4 due to low overhead, energy efficient transmission and no tight synchronization requirement as compared to the beacon-mode transmissions. We model non beacon mode operation of an 802.15.4 unslotted CSMA/CA using Markov chain analysis for low data rate unsaturated traffic. Our modeling is reasonably detailed to capture an exact 802.15.4 system. The modeling captures various network features in detail, such as channel access probability, packet collision, acknowledged transmissions, exponential backoff with retry limits. The packet generation process at the sensor nodes is considered as periodic. We thoroughly derive the packet delay distribution and study the system throughput and energy efficiency trade off which is often crucial for the low rate networks. Our exact Markov chain based analysis reveals that the service process of the protocol is close to an exponential distribution. Therefore, we approximate it by a single server queue with an exponential service time. We then analyze a D/M/1 queue with an unsaturated periodic traffic and show that it can be a good fit for the 802.15.4 system. Using D/M/1 queue model, we derive an end-to-end delay and analyze the delay performance. We also study a system with non-exponential backoff and compare the optimal network performance with the legacy exponential backoff system. Our simulation results corroborate with the theoretical analysis. Our analysis can be used as a system design tool where various network system parameters can be configured to meet a desired end-to-end delay, energy efficiency and throughput.

With the buzz of Internet of Things (IoT), the low power sensor are getting deployed on large scale in applications such as industrial monitoring, smart homes and e-health. For the reason that the sensor devices are battery operated, though the round-robin based Time Division Multiple Access (TDMA) scheduling is the best choice for its high energy and throughput efficiency, the performance degrades as the network scales because of longer frame duration. We address this problem by modelling the network as a tree with interfering links corresponding to the number of devices. We propose a network partitioning approach to reduce these interfering links by extending the tree to a second level. The first level interfering links correspond to optimizing the frame duration where as the second level interfering links correspond to improving the slot utilization. We show that forming this two-level tree is an NP-complete problem by formulating it as a conflict graph. We then propose two polynomial-time algorithms to address offline and online slot allocation requests that resolve these conflicts efficiently by introducing the so called virtual interference links. We illustrate the performance benefits of our proposed algorithms over the conventional round-robin scheduling in terms of improved throughput, scalability and delay.

Performance of a delay-tolerant network has strong dependence on the nodes participating in data transportation. Such networks often face several resource constraints especially related to energy. Energy is consumed not only in data transmission, but also in listening and in several signaling activities. On one hand these activities enhance the system's performance while on the other hand, they consume a significant amount of energy even when they do not involve actual node transmission. Accordingly, in order to use energy efficiently, one may have to limit not only the amount of transmissions, but also the amount of nodes that are active at each time. Therefore, we study two coupled problems: 1) the activation problem that determines when a mobile will turn on in order to receive packets; and 2) the problem of regulating the beaconing. We derive optimal energy management strategies by formulating the problem as an optimal control one, which we then explicitly solve. We also validate our findings through extensive simulations that are based on contact traces.

This note is to correct the author list and authors' affiliations for the above listed paper (ibid., vol. 29, no. 8, pp. 1630-1640, Sep 2011). The acknowledgements are also revised.

The paper establishes a general approach for analyzing queueing models with repeated inhomogeneous vacations. The server goes on for a vacation if the inactivity prolongs more than the vacation trigger duration. Once the system enters in vacation mode, it may continue for several consecutive vacations, possibly with a different probability distribution. We study a simple M/G/1 queue, which has the advantage of being tractable analytically. The theoretical model is applied to the problem of power saving for mobile devices in which the sleep durations of a device correspond to the server vacations. Various system performance metrics such as the frame response time and the economy of energy are derived. A constrained optimization problem is formulated to maximize the economy of energy in power save mode with QoS constraints. An illustration of the proposed methods is shown with a WiMAX system scenario to obtain design parameters for better performance. Our analysis allows us not only to optimize the system parameters for a given traffic intensity but also to propose parameters that provide the best performance under worst case conditions.

We consider a mobile connected to a base station, and study how to optimally schedule shutting off its transceiver. First, we study the model from optimal control perspective. We consider off-times (periods of inactivity) of (controlled) duration. We study the question of scheduling "waking up" instants in which the mobile communicates with the base station and checks whether the inactivity period is over. There is a cost proportional to the delay from the moment the off-time ends until the mobile discovers it, a (small) running cost while the mobile is sleeping and a cost for waking up. We present conditions for optimal sleep periods to be constant and derive the optimal period. For the case that the conditions do not hold, we obtain suboptimal solutions which perform strictly better than the optimal constant one. We then investigate optimality restricted to classes of policies with specific constraints. We adopt the parametric optimization approach which entails cost minimization for a given parameterized policy and selection of the best policy among a class. We then compare the performance of optimal policies, of the proposed suboptimal policies as well as that of standard policies like IEEE 802.16e.

Much research has been devoted to maximize the life time of mobile ad-hoc networks. Life time has often been defined as the time elapsed until the first node is out of battery power. In the context of static networks, this could lead to disconnectivity. In contrast, Delay Tolerant Networks (DTNs) leverage the mobility of relay nodes to compensate for lack of permanent connectivity, and thus enable communication even after some nodes deplete their stored energy. One can thus consider the lifetimes of nodes as some additional parameters that can be controlled to optimize the performance of a DTN. In this paper, we consider two ways in which the energy state of a mobile can be controlled. Both listening and transmission require energy, besides each of these has a different type of effect on the network performance. Therefore we consider a joint optimization problem consisting of: i) activation, which determines when a mobile will turn on in order to receive packets, and ii) transmission control, which regulates the beaconing. The optimal solutions are shown to be of the threshold type. The findings are validated through extensive simulations.

No abstract available.

In this paper, we are concerned with energy efficient area monitoring using information coverage in wireless sensor networks, where collaboration among multiple sensors can enable accurate sensing of a point in a given area-to-monitor even if that point falls outside the physical coverage of all the sensors. We refer to any set of sensors that can collectively sense all points in the entire area-to-monitor as a full area information cover. We first propose a low-complexity heuristic algorithm to obtain full area information covers. Using these covers, we then obtain the optimum schedule for activating the sensing activity of various sensors that maximizes the sensing lifetime. The scheduling of sensor activity using the optimum schedules obtained using the proposed algorithm is shown to achieve significantly longer sensing lifetimes compared to those achieved using physical coverage. Relaxing the full area coverage requirement to a partial area coverage (e.g., 95% of area coverage as adequate instead of 100% area coverage) further enhances the lifetime.

Employing multiple base stations is an attractive approach to enhance the lifetime of wireless sensor networks. In this paper, we address the fundamental question concerning the limits on the network lifetime in sensor networks when multiple base stations are deployed as data sinks. Specifically, we derive upper bounds on the network lifetime when multiple base stations are employed, and obtain optimum locations of the base stations (BSs) that maximize these lifetime bounds. For the case of two BSs, we jointly optimize the BS locations by maximizing the lifetime bound using a genetic algorithm based optimization. Joint optimization for more number of BSs is complex. Hence, for the case of three BSs, we optimize the third BS location using the previously obtained optimum locations of the first two BSs. We also provide simulation results that validate the lifetime bounds and the optimum locations of the BSs.

Increasing network lifetime is important in wireless sensor/ad-hoc networks. In this paper, we are concerned with algorithms to increase network lifetime and amount of data delivered during the lifetime by deploying multiple mobile base stations in the sensor network field. Specifically, we allow multiple mobile base stations to be deployed along the periphery of the sensor network field and develop algorithms to dynamically choose the locations of these base stations so as to improve network lifetime. We propose energy efficient low-complexity algorithms to determine the locations of the base stations; they include i) top-Kmaxalgorithm, ii) maximizing the minimum residual energy (max-min-RE) algorithm, and iii) minimizing the residual energy difference (MinDiff-RE) algorithm. We show that the proposed base stations placement algorithms provide increased network lifetimes and amount of data delivered during the network lifetime compared to single base station scenario as well as multiple static base stations scenario, and close to those obtained by solving an integer linear program (ILP) to determine the locations of the mobile base stations. We also investigate the lifetime gain when an energy aware routing protocol is employed along with multiple base stations