Transfer Learning for Human & AI

Social networks are widely used for information consumption anddissemination, especially during time-critical events such as natural disasters . Despite its significantly large volume, social media content is often too noisy for direct use in any application . In this paper, we present a new large-scale dataset with~77K human-labeled tweets, sampled from a pool of ~24 million tweets across 19disaster events that happened between 2016 and 2019 .…

We describe a novel method for In-Database Machine Learning (IDBML) We contribute a process for . SQL-codegeneration in Python using template macros in Jinja2 as well as the prototype . We measured accuracy and .computation time . Our multidimensional probability estimation was significantlymore accurate than Naive Bayes .…

Reinforcement Learning (RL) is a widely-used technique in many domains, including autonomous driving, robotics, stock trading, and video games . The black box nature of RL agents makes it increasingly important that humans understand the reasoning behind the actions taken by an RL agent, particularly in safety-critical domains .…

Graph search is one of the most successful algorithmic trends in nearneighbor search . We find thatreordering improves the query time by up to 40%, and we demonstrate that the time needed to reorder the graph is negligible compared to the time required to construct the index .…

This paper proposes a method for temporal parallelisation of dynamicprogramming solutions of optimal control problems . We also derive the temporalparallelisation of the linear quadratic tracking control problem . For these twoproblems, we derive the elements and associative operators to be able to use parallel scans to solve these problems with logarithmic time complexity .…

A vast and growing number of IoT applications connect physical devices, suchas scientific instruments, technical equipment, machines, and cameras, across heterogeneous infrastructure from the edge to the cloud . As edge and cloud layers are increasingly tightly integrated, imbalancedresource allocations and sub-optimally placed tasks can quickly deteriorate the overall system performance .…

Persistent memory (PM) is resilient to power loss, provides fast and fine-granular access to memory (unlike disk storage) and has latency and bandwidth close to DRAM (in the order of ns andGB/s) We present PLINIUS, a ML framework using Intel SGXenclaves for secure training of ML models and PM for fault toleranceguarantees .…

This letter studies an unmanned aerial vehicle (UAV) aided multicasting (MC) system . It is enabled by simultaneous free space optics (FSO) backhaul and power transfer . The UAV applies the power-splitting technique to harvestwireless power and decode backhaul information simultaneously over the FSOlink .…

Dynamic GraphConvolutional Network (DyGCN) generalizes the embedding propagation scheme of GCN to dynamicsetting in an efficient manner . The most affected nodes are first updated, and then their changes are propagated to the further nodes and leads to theirupdate . Extensive experiments conducted on various dynamic graphs demonstratethat our model can update the node embeddings in a time-saving andperformance-preserving way .…

NeuMIP is a neural method for representing and rendering a variety of material appearances at different scales . Classical prefiltering(mipmapping) methods fail to generalize to normals, self-shadowing, fibers or more complex microstructures . We also introduce neural offsets, a novel method which allows rendering materials with intricate parallax effects without anytessellation .…

In recent years, there are great interests as well as challenges in applyingreinforcement learning (RL) to recommendation systems . In this paper, wesummarize three key practical challenges of large-scale RL-based recommendersystems: massive state and action spaces, high-variance environment, and theunspecific reward setting in recommendation .…

Multi-access edge computing (MEC) and non-orthogonal multiple access (NOMA) have been regarded as promising technologies to improve computation capability and offloading efficiency of the sixth generation (6G) mobile system . This paper mainly focuses on the hybrid NOMA-MEC system, where multiple users are first grouped into pairs, and users in each pair offload their tasks simultaneously .…

This article is a multiauthored portrait of Edsger Wybe Dijkstra that includes testimonials written by friends, colleagues and students . It provides unique insights into his personality, working style andhabits .…

The ELO ranking system has proven to be a reliable method for calculating therelative skill levels of players in zero-sum games . The evaluation of player strength in trick-taking card games like Skat orBridge is not obvious . We propose a new ELO system for Skat to overcome these weaknesses, based on a tournament scoringsystem .…

We study dynamic matching in a spatial setting . Riders arrive in some (possibly adversarial) order at random points . Unmatched riders incur a waiting cost $c$ per period . The cost of matching a driver to a rider is equal to the distance between them .…

Scientific knowledge cannot be seen as a set of isolated fields, but as ahighly connected network . Understanding how research areas are connected is ofparamount importance for adequately allocating funding and human resources . Relationshipbetween disciplines can be drawn from data on data on the trajectory of individualscientists, as researchers often make contributions in a small set of interrelated areas .…

We propose a novel construction for the well-known prefer-max De Bruijnsequence . We show that the construction implies known results from the literature in a straightforward manner . It also forms analternative proof for the seminal FKM-theorem . We also show that it implies the correctness of onion-theoracle onion theorem and of the shift rules for preferred-max and prefer-min .…

LI-Net is a large-pose Identity-preserving face reenactment network . The Landmark Transformer is adopted to adjust driving landmark images, which narrows the identity gap between driving and source landmark images . Then the Face Rotation Module and the Expression Enhancing Generator decouple the transformed landmark image into pose and expression features, and re-enact those attributes separately to generate identity-preserved faces with accurate expressions and poses .…

Audio-visual event localization aims to localize an event that is bothaudible and visible in the wild . This task is a widespread audio-visual sceneanalysis task for unconstrained videos . We propose aMultimodal Parallel Network (MPN) which can perceive global semantics and local information parallelly .…

Federated Learning (FL) allows edge devices to collaboratively learn a shared prediction model while keeping their training data on the device . We present an exploration of on-device FL on various smartphones andembedded devices using the Flower framework . We also evaluate the system cost and discuss how this quantification could be used to design more efficient FL algorithms .…

Eigen vector perturbation analysis plays a vital role in various statistical data science applications . The proposed estimators are nearly minimax optimaleven when the associated eigen-gap is substantially smaller than what is required in prior theory . In order to mitigate a non-negligible bias issue inherent to thenatural “plug-in” estimator, we develop de-biased estimators that (1) achieveminimax lower bounds for a family of scenarios (modulo some logarithmicfactor) and (2) can be computed in a data-driven manner without samplesplitting.…

Scaling blockchain efficiency is crucial to its widespread usage in which thepayment channel is one of the most prominent approaches . With payment channelsand the network they construct, two users can move some transactions off-chain in a predetermined duration to avoid expensive and time-consuming on-chainsettlements .…

The traditional Newton-Raphson method suffers from non-convergence issues when solving post-outage power flow problems . The HE-based simulator guarantees convergence if the true power flow solution exists . The PHE method does notundermine the numerical robustness of HE and significantly reduces thecomputation burden in large-scale contingency analysis .…

In this paper, we study the optimal resource allocation algorithm design for large intelligent reflecting surface (IRS)-assisted simultaneous wirelessinformation and power transfer (SWIPT) systems . We aim tominimize the total base station (BS) transmit power by jointly optimizing the beamforming and the transmission mode selection policy taking into account the quality-of-service requirements of information decoding and non-linear energyharvesting receivers .…

We study the computational complexity of $c$-Colored $P_\ell$ Deletion and $c$. Deletions . We show that both problems are fixed-parameter tractable with respect to the colored neighborhood diversity of the input graph . We also provide hardness results to outline the limits ofparameterization by the standard parameter solution size $k$.…

Counting the number of nodes in Anonymous Dynamic Networks is enticing from an algorithmic perspective . Counting cannot be solved deterministically without at least one distinguished node . We present negative and positive results to answer these questions . To the best of our knowledge, this is the best work that addresses them.…

Smart Cities and Communities (SCC) constitute a new paradigm in urbandevelopment . SCC ideates on a data-centered society aiming at improving efficiency by automating and optimizing activities and utilities . Thispaper describes AI perspectives in SCC and gives an overview of AI-basedtechnologies used in traffic to enable road vehicle automation and smarttraffic control .…

Java is the “go-to” programming language choice for developing scalableenterprise cloud applications . In such systems, even a few percent CPU timesavings can offer a significant competitive advantage . In this paper, we present DJXPerf, a lightweight, object-centric memoryprofiler for Java, which associates memory-hierarchy performance metrics (e.g.,…

Universal register machine, a formal model of computation, can be emulated on the array of the Game of Life, a two-dimensional cellular automaton . We performspectral analysis on the computation dynamical process of the universalregister machine . The array is divided into small sectors and the power spectrum is calculated from the evolution in each sector .…

Human multi-robot system (MRS) collaboration is demonstrating potentials in wide application scenarios . A novel Synthesized Trust Learning (STL) method was developed to model human trust in the collaboration . STL explores two aspects of human trust (trust level and trust preference) and accelerates convergence speed by integrating active learning to reduce workload .…

In this paper, we propose a quadruple-UPA architecture for realizing the beamforming with wide coverage in terahertz (THz) communication . We propose a 3D beam training procedure to find the optimal beamforming solution, i.e., the optimal narrow-beam pair, in a fast manner .…

We introduce a non-parametric approach for infinite video texture synthesis using a representation learned via contrastive learning . We learnrepresentations for video frames and frame-to-frame transition probabilities byfitting a video-specific model . Tosynthesize a texture, we randomly sample frames with high transitionprobabilities to generate diverse temporally smooth videos with novel sequencesand transitions .…

Social media is both helpful and harmful to the work against sex trafficking . The thesis explores the use of semi-supervised classification to identify social media posts that are part of the sex industry . AI and Machine Learning (ML) have been used in work against sextrafficking, but they predominantly focus on detecting Child Sexual Abuse Material .…

In this paper, we study community detection when we observe $m$ sparsenetworks and a high dimensional covariate matrix, all encoding the same community structure among $n$ subjects . In the asymptotic regime where thenumber of features and the number of subjects $p$ grows proportionally, we find a sharp threshold of phase transition .…

Advances on resource disaggregation will enable transparency for most Cloud applications . ServerlessSuperComputer or Infinite Laptop proposals will be equivalent to Tim Wagner’s serverless supercomputer proposals . We present five serverless predictions for the next decade that will realize this vision of transparency .…

Power ISA(TM) Version 3.1 has introduced a new family of matrix mathinstructions, collectively known as the Matrix-Multiply Assist (MMA) facility . These instructions have led to a power- and area-efficient implementation of ahigh throughput math engine in the future POWER10 processor .…

In this paper we analyze the problem of optimal task scheduling for datacenters . We propose a fast distributediterative algorithm which operates over a large scale network of nodes . The algorithm is guaranteed to converge to the exact optimal scheduling plan in a finitenumber of time steps .…

We proposePilot-Edge as a common abstraction for resource management across the edge-to-cloud continuum . Pilot-Edge is based on the pilot abstraction, which provides aFunction-as-a-Service (FaaS) interface for application-level tasks . Theabstraction allows applications to encapsulate common functions in high-leveltasks that can then be configured and deployed across the continuum .…

Presenting two new Petri net slicing algorithms to isolate those places and transitions that may contribute tokens to one or more places given (the slicing criterion) The completeness of thefirst algorithm and the minimality of the second algorithm are formally proven .…

This paper proposes two novel approaches to perform more suitable sensitivity analyses for discrete topology optimization methods . We introduce a more formal description of the Bi-directional Evolutionary EvolutionaryStructural Optimization (BESO) method . The novel Woodbury approach provides exactsensitivity values and is a better alternative to HOCI.…

PrivGenDB is the first SSE-based approach ensuring the confidentiality of shared SingleNucleotide Polymorphism (SNP)-phenotype data through encryption . It supports a variety of query types on genomic data, including countqueries, Boolean queries, and k’-out-of-k match queries . Computer evaluations on a dataset with 5,000 records and 1,000 SNPs demonstrate that a count/Boolean query and a k-out-to-match query over 40 SNPs take approximately 4.3s and 86.4{\mu}s, respectively, that outperforms the existing schemes .…

Electronic Health Record (EHR) data has been of tremendous utility inArtificial Intelligence for healthcare such as predicting future clinicalevents . These tasks often come with many challenges when using machine learning models due to a myriad of factors including classimbalance and data heterogeneity .…

The task of calculating similarities between strings held by different organizations without revealing these strings is an increasingly important problem in areas such as health informatics, national censuses, genomics, and fraud detection . We incorporate a hashing basedencoding technique upon the encoded suffixes to improve privacy againstfrequency attacks such as those exploiting Benford’s law .…

Using anempirical design analysis, we identify interface interferences impacting users’ online privacy . We find usage behaviour changes due to increased privacy concerns and report individual cases of addiction and mental health issues . These observations are the results of arapidly changing SNS creating a gap of understanding between users’ interactions with the platform and future consequences .…

TenSEAL is an open-source library for Privacy-Preserving Machine Learning usingHomomorphic Encryption that can be easily integrated within popular machinelearning frameworks . We benchmark our implementation using MNIST and show that an encrypted convolutional neural network can be evaluated in less than asecond, using less than half a megabyte of communication.…

Weakly-supervised temporal action localization aims to localize actioninstances temporal boundary and identify the corresponding action category with only video-level labels . Traditional methods mainly focus on foreground and background frames separation with only a single attention branch and classactivation sequence .…

In many information extraction applications, entity linking (EL) has emerged as a crucial task that allows leveraging information about named entities from a knowledge base . In this paper, we address the task of multimodal entitylinking (MEL), an emerging research field in which textual and visualinformation is used to map an ambiguous mention to an entity in a knowledgebase .…

Terahertz (THz) communications have emerged as a promising candidate to support the heavy data traffic and exploding network capacity in the future 6G networks . However, THz communications are facing many challenges forpractical implementation, such as propagation loss, signal blockage, and hardware cost .…

A k-anonymous broadcast can be implemented using a small group of diningcryptographers to first share the message, followed by a flooding phase started by group members . Members have little incentive to forward the message in atimely manner, as forwarding incurs costs, or they may even profit from keeping the message .…

In this paper, we study linear spaces of matrices defined over discretelyvalued fields . We take first steps into the theory ofrank-metric codes over discrete valuation rings by means of skew algebras derived from Galois extensions of rings . We modelprojectivizations of rank-metic codes via Mustafin varieties, which we then give sufficient conditions for a decrease in the dimension .…