Transfer Learning for Human & AI

Presentations are critical for communication in all areas of our lives, yet the creation of slide decks is often tedious and time-consuming . We present D2S, a novel system that tackles the document-to-slides task with a two-step approach: 1)Use slide titles to retrieve relevant and engaging text, figures, and tables;2) Summarize the retrieved context into bullet points with long-form questionanswering.…

In this paper, we consider a reconfigurable intelligent surface(RIS)-assisted two-way relay network . We propose two algorithms to design the RIS phase shifts and the BSpower amplification parameter, namely the SNR-upper-bound-maximization (SUM)method, and genetic-SNR (GSM) method . Our objective is tomaximize the minimum signal-to-noise ratio (SNR) of the two users, under the transmit power constraint at the BS .…

Zero-error single-channel source codes have no advantage oversingle-channel codes in data compression, making them worthless in most applications . We propose a suboptimal code construction whose redundancy is guaranteed to be no larger than that of an optimal single channel sourcecode .…

In this letter, we consider an underlay cognitive radio network assisted bydual-hop decode-and-forward (DF) relaying . For a general multi-user network, we adopt a max-min fairness relay selection scheme and analyse the outageprobability when the channels are subject to independent and non-identicalNakagami-m fading .…

Studies related to fairness in machine learning have recently gained traction due to its ever-expanding role in high-stakes decision making . Previously, these problems have been studied under asetting where sensitive attributes, with respect to which fairness conditions impose diversity across clusters, are assumed to be observable .…

Matrix factorization is an important representation learning algorithm, e.g.,recommender systems, where a large matrix can be factorized into the product of two low dimensional matrices termed as latent representations . This paper investigates the problem of matrix factorization in distributed computingsystems with stragglers, those compute nodes that are slow to returncomputation results .…

Intent modelling has become an important part of modern dialogue systems . Up until recently the focus has been on detecting a fixed,discrete, number of seen intents . This paper proposes a novel model where intents are continuous points placed in a specialist Intent Space that yields severaladvantages.…

Recent advances in Named Entity Recognition show that document-levelcontexts can significantly improve model performance . In many applicationscenarios, however, such contexts are not available . In this paper, we proposeeto find external contexts of a sentence by retrieving and selecting a set ofsemantically relevant texts through a search engine .…

We propose a user satisfaction annotation dataset, USS, that includes 6,800 dialogues sampled from multiple domains . All user utterances in thosedialogues, as well as dialogues themselves, have been labeled based on a5-level satisfaction scale . We also share three baseline methods for usersatisfaction prediction and action prediction tasks .…

Recently, neural architecture search (NAS) has attracted much attention and has been explored for automatic speech recognition (ASR) In thiswork, we focus on streaming ASR scenarios and propose the latency-controlledNAS for acoustic modeling . Extensive experiments show that: Based on the proposed neuralarchitecture, the neural networks with a medium latency of 550ms (millisecond)and a low latency of 190ms can be learned in the vanilla and revised operationspace respectively .…

We propose a novel personalized speechenhancement method to adapt a compact denoising model to the test-timespecificity . This zero-shot learning procedure circumvents the process of collecting users’ clean speech, which users are reluctant to comply due toprivacy concerns and technical difficulty of recording clean voice .…

Edge computing as a promising technology provides lower latency, moreefficient transmission, and faster speed of data processing . Each edge server with limited resources can offload latency-sensitive and computation-intensive tasks from nearby user devices . Edge computing faces challenges such as resource allocation, energy consumption, security and privacy issues .…

A flip-swap language is a set S of binary strings of length n such that $S\cup 0^n$ is closed under two operations (when applicable): (1) Flip theleftmost 1; and (2) Swap the leftmost 1 with the bit to its right .…

We introduce a new problem which we call the Pony Express problem . n robots with differing speeds are situated over some domain . The objective is to deliver the message in minimum time . We provide an offline algorithm running in O(n log n) running time and an online algorithm that attains a competitive ratio of 3/2 which we show is the best possible .…

Data movement between the CPU and main memory is a first-order obstacle against improving performance, scalability, and energy efficiency in modern systems . We develop the first systematic methodology to classify applications based on the sources contributing to data movement bottlenecks .…

LPVcore software package for MATLAB developed to model, simulate, estimate and control systems via linear parameter-varying(LPV) input-output (IO), state-space (SS) and linear fractional (LFR) representations . The paper contains an overview of functions in the toolbox to simulate and identify and identify IO, SS and LFR representations .…

In the rendezvous problem, two computing entities (called \emph{agents) have to meet at the same vertices in a graph . The only known result is that the time complexity of $O(\sqrt{n)$ rounds is achievable if the graph is complete and agents areprobabilistic, asymmetric, and can use whiteboards placed at vertices .…

We build on abduction-based explanations for ma-chine learning and develop amethod for computing local explanations for neural network models in naturallanguage processing . Our explanations comprise a subset of the words of the in-put text that satisfies two key features: optimality w.r.t.…

The main findings of this article will provide an overview in amethodology perspective to study theoretical model for blockchain fundamentals and design network service for blockchain-based mechanisms and algorithms . Main findings are to provide overview of theoretical model of blockchain fundamentals .…

We initiate the systematic study of QMA algorithms in the setting of propertytesting, to which we refer as QMA proofs of proximity (QMAPs) We show quantum speedups for properties that lie outside of thisfamily, such as graph bipartitneness . We also investigate the complexitylandscape of this model, showing that QMAPs can be exponentially stronger than both classical and quantum testers .…

Jump functions are the most studied non-unimodal benchmark in the theory of randomized search heuristics, in particular, evolutionary algorithms . They have significantly improved our understanding of how EAs escape from localoptima . To leave the local optimum one can only jump directly to the global optimum .…

Radio fingerprinting (RFP) is a compelling approach that leverages the imperfections in the tag’s wireless circuitry to achieve large-scale RFID clone detection . Time-varying channel conditions can significantly decrease the accuracy of the RFP process . We propose a novel training framework based on federatedmachine learning (FML) and data augmentation (DAG) to boost the accuracy .…

Oxygen Depletion problem is an implicit free boundary value problem . We show severalmathematical formulations of this model from the literature and give a newformulation based on a gradient flow with constraint . Weshow a convergence result for an approximation based on an approximation is shown .…

Vehicular Ad hoc Network (VANET) is a new sort of wireless ad-hoc network . Routing data packets is challenging due to frequent changes of network topology because of the highly dynamic nature of vehicles . Many position-based routing protocols have been developed for routing messages that have been identified to be appropriate for VANETs.…

Recent advances in neural architectures have revived the problem of morphological rule learning . We evaluate the Transformer as a model of morphology rule learning and compare it with Recurrent Neural Networks (RNN) on English, German, and Russian . We bring to the fore a hitherto overlookedproblem, the morphological gaps, where the expected inflection of a word is missing .…

We develop new algorithmic methods with provable guarantees for featureselection in regard to categorical data clustering . Our algorithm is based on asolution to a more general problem, Constrained Clustering with Outliers . We assume that there are some inadvertent (or undesirable)features of the input data that unnecessarily increase the cost of clustering on the selected features .…

The notions of synchronizing and partitioning sets are recently introducedvariants of locally consistent parsings with great potential inproblem-solving . In this paper we propose a deterministic algorithm thatconstructs for a given readonly string of length $n$ over the alphabet$\{0,1,\ldots,n^{\mathcal{O}(1)}\}$ The algorithm uses a version of a set withsize $tau = \frac{n/b$ using $taus$ and $\tau$-partitioning set .…

Next generation mobile networks need to expand towards uncharted territories in order to enable the digital transformation of society . In this context,aerial devices such as unmanned aerial vehicles (UAVs) are expected to addressthere gap in hard-to-reach locations . But limited battery-life is an obstacle for the successful spread of such solutions .…

Hibridizable Discontinuous Galerkin discretizations for a class of non-linear interior elliptic boundary value problems . We show that, under mild assumptions on thesource term and the computational domain, the discrete systems are well posed. We provide a priori error estimates showing that the discretesolution will have optimal order of convergence as long as the distance betweenthe curved boundary and the computation boundary remains of the same order ofmagnitude as the mesh parameter .…

Physics-based simulation for fluid flow in porous media is a computational technology to predict the temporal-spatial evolution of state variables . This letter describes adeep learning (DL) workflow to predict pressure evolution as fluid flows in large-scale 3D heterogeneous porous media .…

AnomalyHop is mathematically transparent, easy to train, and fast in its inference speed . Its area under the ROC curve (ROC-AUC)performance on the MVTec AD dataset is 95.9%, which is among the best of several benchmarking methods . Comparing with state-of-the-art image anomaly localization methods based on deep neuralnetworks (DNNs) The code is publicly available at Github and available in the works .…

In this paper, we introduce a definition of phase response for a class of MIMO linear time-invariant systems whosefrequency responses are (semi-)sectorial at all frequencies . The newly definedphase concept subsumes the well-known notions of positive real systems and negative imaginary systems .…

We introduce a novel class of time integrators for dispersive equations . They reproduce the dynamics of the solution from the classical $varepsilon = 1$ up to long wave limit regime . The new schemes converge with rates at order $t= \frac{1$ overlong times .…

Negation is a core construction in natural language . State-of-the-art pre-trained language models often handle negation incorrectly . We propose to augment the language modeling objective with an unlikelihood objective that is based on negated generic sentences from a raw text corpus .…

Alarm root cause analysis is critical for efficient and accurate fault localization and failure recovery . We propose a novel data-driven framework for root cause alarm localization, combining both causal inference and network embedding techniques . We subsequently discover root cause alarms in a real-time data stream by applying an influence maximization algorithm on the weighted graph .…

This paper considers the problem of retrieving an object from a set oftightly packed objects by using a combination of robotic pushing and grasping actions . The proposed solution, Visual Foresight Tree (VFT),cleverly rearranges the clutter surrounding the target object so that it can begrasped easily .…

Roboticists frequently turn to Imitation learning (IL) for data efficient policy learning . Many IL methods combat distributional shift issues with older BehaviorCloning (BC) methods by introducing oracle experts . We present adata-efficient imitation learning technique called Collocation forDemonstration Encoding (CoDE) that operates on only a fixed set of trajectorydemonstrations by modeling learning as empirical risk minimization .…

The trustworthiness of Robots and Autonomous Systems (RAS) has gained aprominent position on many research agendas towards fully autonomous systems . RAS must be (i)safe in any uncertain and dynamic surrounding environments; (ii) secure, thusprotecting itself from any cyber-threats; (iii) healthy with fault tolerance;(iv) trusted and easy to use to allow effective human-machine interaction(HMI) The risks posed by advanced AI in RAS have not received sufficient scientific attention .…

Visual localization techniques often comprise a hierarchical localizationpipeline, with a visual place recognition module used as a coarse localizer toinitialize a pose refinement stage . We demonstrate significant improvements to the localizationaccuracy of the coarse localization stage using our methods, whilst retaining state-of-the-art performance under severe appearance change .…

Current Category-levelArticulation Pose Estimation (CAPE) methods are studied under the single-instance setting with a fixed kinematic structure for each category . We reform this problem setting for real-world environments and suggest a CAPE-Real (CAPER) task setting . This setting allows multiple instancesto co-exist in an observation of real world .…

In this paper, we propose a tightly-coupled, multi-modal simultaneouslocalization and mapping (SLAM) framework, integrating an extensive set of sensors: IMU, cameras, multiple lidars, and Ultra-wideband (UWB) rangemeasurements . We propose a two stage approach . In the first stage,lidar, camera, and IMU data on a local sliding window are processed in a coreodometry thread .…

In this paper the reversibility of executable Interval Temporal Logic (ITL)specifications is investigated . ITL allows for the reasoning about systems interms of behaviours which are represented as non-empty sequences of states . At a high level one can specify a system in terms of properties, for instance safetyand liveness properties .…

Biologically inspired algorithms for simultaneous localization and mapping (SLAM) such as RatSLAM have been shown to yield effective and robust robotnavigation in both indoor and outdoor environments . One drawback however is the sensitivity to perceptual aliasing due to the template matching of low-dimensional sensory templates .…

Past and current research has typically focused on ensuring that searchtechnology for the classroom serves children . In this paper, we argue for theneed to broaden the research focus to include teachers and how searchtechnology can aid them . We share how furnishing abehind-the-scenes portal for teachers can empower them by providing a window into the spelling, writing, and concept connection skills of their students .…

Report: fAshIon (artificial intelligence (AI) in fashion) explores its potentiality to become a major disruptor of the fashion industry in the near future . Report: Many potential opportunities exist for the use of AI in fashion which can transform the industry embedded with AItechnologies and boost profits .…

Sobi is a mobile service robot developed as an interactive guide for open environments such as public places with indoor and outdoor areas . The robot will serve as a platform for environmental modeling and human-robot interaction . Sobi patrols a total of 66.6 km with an average of 5.5 hours of travel time per weekday, chargingautonomously in between .…

Current algorithms for constructing sound and precise call graphsanalyze complex program dependencies, therefore they might be difficult toscale . We propose NoCFG, a new sound and scalable method for approximating a call graph that supports a wide variety of programming languages .…

“Spider” is a nickname of *special Frobenius algebras, a fundamental structure from mathematics, physics, and computer science . Pregroups and spiders have been used in natural language processing: one for syntax, the other forsemantics . The compositional framework that emerged with the results suggests new ways tounderstand and apply the basis structures in machine learning and data analysis .…

Learning prerequisite chains is an essential task for efficiently acquiring knowledge in both known and unknown domains . In this paper, we propose unsupervised cross-domain conceptprerequisite chain learning using an optimized variational graph autoencoder . The annotated data and resources, as well as the code, will be made publiclyavailable .…