LinkLouvain Link Aware A B Testing and Its Application on Online Marketing Campaign

Theaverage treatment effect (ATE) of campaign strategies need to be monitored throughout the campaign . A/B testing is usually conducted for such needs, but the existence of user interaction can introduce interference to normal testing . With the help of link prediction, LinkLouvain design a way to minimize graph interference and it gives an accurate andsound estimate of the campaign’s ATE .…

Temporal Motifs in Smart Grid

The energy consumptionpattern across the appliances, houses, communities and entire cities help energy utility companies and consumers plan their electricity generation and consumption . The edge or connection represents energy flow between two participants of the network, these connections last till the power is being consumed/generated .…

Agent Incentives A Causal Perspective

We present a framework for analysing agent incentives using causal influencediagrams . We establish that a well-known criterion for value of information iscomplete . We propose a new graphical criterion for the value of control,establishing its soundness and completeness . We also introduce two new concepts: response incentives indicate which changes in theenvironment affect an optimal decision, while instrumental control incentives establish whether an agent can influence its utility via a variable X .…

Towards Multi agent Reinforcement Learning for Wireless Network Protocol Synthesis

This paper proposes a multi-agent reinforcement learning based medium accessframework for wireless networks . The access problem is formulated as a MarkovDecision Process (MDP), and solved using reinforcement learning with every network node acting as a distributed learning agent . It is shown that by learning to adjust MAC layer transmissionprobabilities, the protocol is not only able to attain theoretical maximumthroughput at an optimal load, but unlike classical approaches, it can alsoretain that maximum throughput at higher loading conditions .…

Approximately Solving Mean Field Games via Entropy Regularized Deep Reinforcement Learning

The recent mean field game (MFG) formalism facilitates otherwise intractablecomputation of approximate Nash equilibria in many-agent settings . We show that all discrete-time finite MFGs with non-constant fixedpoint operators fail to be contractive as typically assumed in existing MFGliterature . Instead, we incorporate entropy-regularization and Boltzmann policies into the fixed pointiteration .…

Exploiting Raw Images for Real Scene Super Resolution

Super-resolution is a fundamental problem in computer vision which aims toovercome the spatial limitation of camera sensors . Most algorithms only perform well on synthetic data, which limits their applications in real scenarios . We propose a method to generate more realistictictraining data by mimicking the imaging process of digital cameras .…

On Robustness of Neural Semantic Parsers

Semantic parsing maps natural language (NL) utterances into logical forms (LFs) underpins many advanced NLP problems . Semantic parsers gain performance boosts with deep neural networks, but inherit vulnerabilities against adversarial examples . A scalable methodology is proposed to construct robustness testsets based on existing benchmark corpora.…

Metrics and continuity in reinforcement learning

In most practical applications of reinforcement learning, it is untenable tomaintain direct estimates for individual states . Instead, researchers often leverage state similarity to build models that can generalize well from alimited set of samples . The notion of state similarity used, and theneighbourhoods and topologies they induce, is crucial importance .…

An Abstraction based Method to Verify Multi Agent Deep Reinforcement Learning Behaviours

Multi-agent reinforcement learning (RL) often struggles to ensure the safebehaviours of learning agents . The approach we propose expresses the constraints to verify inProbabilistic Computation Tree Logic (PCTL) and builds an abstractrepresentation of the system . The abstract model allows for model checking techniques to identify a set ofabstract policies that meet the safety constraints expressed in PCTL .…

Graph Classification Based on Skeleton and Component Features

GraphCSC is a novel graph embedding algorithm that realizes classification based on skeleton information using fixed-order structures learned in anonymous random walksmanner . Two graphs are similar if their skeletons and components are both similar, thus in our model, we integrate both of them together into embeddings as graph homogeneitycharacterization .…

Unassisted Noise Reduction of Chemical Reaction Data Sets

Existing deep learning models applied to reaction prediction in organicchemistry can reach high levels of accuracy (90% for Natural LanguageProcessing-based ones). With no chemical knowledge embedded than theinformation learnt from reaction data, the quality of the data sets plays acrucial role in the performance of the prediction models .…

Graph Coarsening with Neural Networks

Graph coarsening is one popular technique to reduce the size of agraph while maintaining essential properties . It generalizes to graphs of larger size ($25\times$ of training graphs), is adaptive to different losses (differentiable andnon-differentiable) and scales to much larger graphs than previous work .…

Modular approach to data preprocessing in ALOHA and application to a smart industry use case

The paper addresses a modular approach, integrated into the ALOHA tool flow, to support the datapreprocessing and transformation pipeline . This is realized throughcustomizable plugins and allows the easy extension of the tool flow toencompass new use cases . To demonstrate the effectiveness of the approach, we present some experimental results related to a keyword spotting use case and weoutline possible extensions to different use cases.…

QoS Aware Power Minimization of Distributed Many Core Servers using Transfer Q Learning

AQoS-aware runtime controller using horizontal scaling (node allocation) and vertical scaling (resource allocation within nodes) methods synergistically toprovide adaptation to workloads while minimizing the power consumption under QoS constraint . TransferQ-learning, which further tunes power/performance based on workload profile, reduces exploration time and QoS violations compared to model-free Q-learning .…

Fast Exploration of Weight Sharing Opportunities for CNN Compression

The computational workload involved in Convolutional Neural Networks (CNNs) is typically out of reach for low-power embedded devices . There are a largenumber of approximation techniques to address this problem . These methods havehyper-parameters that need to be optimized for each CNNs using design spaceexploration (DSE) The goal of this work is to demonstrate that the DSE phasetime can easily explode for state of the art CNN .…

Neural Data Augmentation via Example Extrapolation

In many applications of machine learning, certain categories of examples maybe underrepresented in training data, causing systems to underperform on such “few-shot” cases at test time . We propose a data augmentation approach that performs neural ExampleExtrapolation (Ex2) Given a handful of exemplars sampled from somedistribution, Ex2 synthesizes new examples that also belong to the samedistribution .…

Anomaly Detection of Time Series with Smoothness Inducing Sequential Variational Auto Encoder

Deep generative models have demonstrated their effectiveness in learninglatent representation and modeling complex dependencies of time series . In thispaper, we present a Smoothness-Inducing Sequential Variational Auto-Encoder(SISVAE) model for robust estimation and anomaly detection of multi-dimensionaltime series . The proposed prior works as a regularizer that places penalty atnon-smooth reconstructions.…

Gaze based dual resolution deep imitation learning for high precision dexterous robot manipulation

A deep imitationlearning based method, inspired by the gaze-based dual resolution visuomotorcontrol system in humans, can solve the needle threading task . The proposed method enables precisemanipulation tasks using a general-purpose robot manipulator and improvescomputational efficiency . The experimental resultsobtained in this study demonstrate that the proposed method enabled .…

Detection of Racial Bias from Physiological Responses

This paper investigates whether we can reliably detect racial bias from physiological responses from heartrate, conductive skin response, skin temperature, and micro-body movements . Our machine learning and statistical analysis show that implicit bias can be predicted from physiological signals with 76.1% accuracy .…

Evaluating the Interpretability of Generative Models by Interactive Reconstruction

Many have argued that machine learning models must be human-interpretable . However, despite increasing interest in interpretability, there remains no firm consensus on how to measure it . This is especially true in representationlearning, where interpretability research has focused on “disentanglement” measures only applicable to synthetic datasets and not grounded in human factors .…

CTC based Compression for Direct Speech Translation

Previous studies demonstrated that a dynamic phone-informed compression of the input audio is beneficial for speech translation . We exploit the Connectionist Temporal Classification (CTC) to compress the input sequence according to its phonetic characteristics . Our experiments demonstrate that our solution brings a 1.3-1.5 BLEU improvement over a strong baseline on two language pairs (English-Italian and English-German), contextually reducing the memory footprint by more than 10% .…

MAUVE Human Machine Divergence Curves for Evaluating Open Ended Text Generation

MAUVE — ametric for open-ended text generation — compares the distribution of machine-generated text to that of human language . The metric measures the meanarea under the divergence curve for the two distributions, exploring the trade-off between two types of errors: those arising from parts of the humandistribution that the model distribution approximates well, and those it doesn’t .…

An Improved Baseline for Sentence level Relation Extraction

Sentence-level relation extraction (RE) aims at identifying the relationship between two entities in a sentence . In this paper, we revisit two aspects of RE models that are not studied, namely entity representation and NA instance prediction . Our improved baseline model, incorporated with entity representations with typemarkers and confidence-based classification for enhanced NA instance detection, significantly outperforms previous SOTA methods .…

The impact of external innovation on new drug approvals A retrospective analysis

Pharmaceutical companies are relying more often on external sources of innovation to boost their discovery research productivity . We analyzed the pre-approval publicationhistories for FDA-approved new molecular entities (NMEs) and new biologicentities (NBEs) launched by 13 top research pharma companies during the lastdecade (2006-2016) We found academic institutions contributed themajority of publications and that publication subject matter isclosely aligned with the strengths of the respective innovator .…

TAPInspector Safety and Liveness Verification of Concurrent Trigger Action IoT Systems

Trigger-action programming (TAP) is a popular end-user programming framework that can simplify the Internet of Things (IoT) automation with simple trigger-action rules . TAPInspector is a novel system to detect vulnerabilities in concurrentTAP-based IoT systems using model checking . It automatically extracts TAP rules from IoT apps, translates them into a hybrid model with model slicing and statecompression, and performs model checking with various safety and livenessproperties .…