Transfer Learning for Human & AI

Humans have mental models that allow them to plan, experiment, and reason inthe physical world . How should an intelligent agent go about learning such models? In this paper, we will study if models of the world learned in anopen-ended physics environment, without any specific tasks, can be reused for downstream physics reasoning tasks .…

Conversational Recommendation System (CRS) uses theinteractive form of the dialogue systems to solve the intrinsic problems of traditional recommendation systems . The existing CRS models are unable to deal with theexploitation and exploration (E&E) problem well, resulting in the heavy burdenon users .…

We present and review an algorithmic and theoretical framework for improving neural network architecture design via momentum . We consider how momentum can improve the architecture design for recurrent neural networks, neural ordinary differential equations (ODEs), and transformers . Weshow that integrating momentum into neural network architectures has several remarkable theoretical and empirical benefits, including overcoming the vanishing gradient issues in training RNNs and neural ODEs, resulting in effective learning long-termdependencies .…

Kraken + Ciaconnareaches a substantially lower character error rate than Tesseract/OCR-Don commentary sections with high density of polytonic Greek text (average CER7% vs. 13%) Tesseracts is slightly more accurate than Kraken +Ciaconna on text sections written predominantly in Latin script .…

We introduce a framework for multi-camera 3D object detection . Our method manipulates predictions directly in 3D space . We achievestate-of-the-art performance on the nuScenes autonomous driving benchmark . Ourarchitecture extracts 2D features from multiple camera images and then uses as .…

The key to personalized news recommendation is to match the user’s interestswith the candidate news precisely and efficiently . With only five historical news selected, it can significantly improve the AUC by 2.17% over the state-of-the-art interaction-based models . SFI can quickly select informative historical newsw.r.t.…

Gaussian boson sampling is a model of photonic quantum computing that has attracted attention as a platform for building quantum devices capable of performing tasks that are out of reach for classical devices . We show that, under the standard anti-concentration and Permanent-of-Gaussians conjectures, there is noefficient classical algorithm to sample from ideal Gaussian Boson samplingdistributions (even approximately) unless the polynomial hierarchy collapses .…

Few-shot class incremental learning is a key challenge for machine learning systems deployed innon-stationary environments . The key to this approach is a new family of subspace regularizationschemes that encourage weight vectors for new classes to lie close to thesubspace spanned by the weights of existing classes .…

We present a novel point-based, differentiable neural rendering pipeline for scene refinement and novel view synthesis . The point cloud rendering is performed by adifferentiable renderer using multi-resolution point rasterization . After rendering, the neural image pyramid is passed through a deep neural network for shading calculations and hole-filling .…

The class of row monomial matrices is closed under multiplication, but not under ordinary matrix addition . The most significant difference is the summation operation . The algebra plays an important role in the study of DFA,especially for synchronizing automata .…

The ELUE benchmark is publicly available athttp://eluebenchmark.fastnlp.top/. We demonstrate the ElasticBERT outperforms or performs on par with SOTA compressed and early exiting models . We alsopre-train and release a strong baseline, ElasticBERt, whose elasticity is bothstatic and dynamic . The benchmark is public and a public leaderboard for forefficient NLP models .…

Magnetic polarizability tensor (MPT) offers an economical characterisation of metallic objects . MPT spectral signature can be determined frommeasurements of the induced voltage over a range frequencies in a metalsignature for a hidden object . With classification in mind, it can also becomomomputed in advance for different threat and non-threat objects .…

We show analytically using a macrospin approximation that easy-plane spinHall nano-oscillators excited by a spin-current have phase dynamics analogous to that of Josephson junctions . We simulate two elementary neuralnetwork blocks that implement operations essential for neuromorphic computing . First, we show that output spikes energies from two neurons can be summed and injected into a following layer neuron and second, we demonstrate that outputspikes can be multiplied by synaptic weights implemented by locally modifying theanisotropy .…

Continuous blood pressure (BP) measurements can reflect a bodys response todiseases and serve as a predictor of cardiovascular and other healthconditions . Current cuff-based BP measurement methods are incapable of providing continuous BP readings . The proposed clustering approach helps obtain more accurate estimates ofSystolic Blood pressure (SBP) and Diastolic Blood Pressure (DBP) Using the clustering .…

We analyse an energy minimisation problem recently proposed for modellingsmectic-A liquid crystals . The optimality conditions give a coupled nonlinearsystem of partial differential equations . Our two main results are aproof of the existence of solutions to the minimisation problems and a priori error estimates for its discretisation using the $\mathcal{C}^0$ interior penalty method .…

This paper concerns solving the steady radiative transfer equation using the physics informed neural networks (PINNs) The ideaof PINNs is to minimize a least-square loss function, that consists of theresidual from the governing equation, the mismatch from the boundaryconditions, and other physical constraints such as conservation .…

Despite widespread deployment of differential privacy, relatively little is known about what users think of it . We find that users care about the kinds of information leaks against which differential privacy protects and are more willing to share their private information when the risks of these leaks are less likely to happen .…

3D object detection often involves complicated training and testingpipelines, which require substantial domain knowledge about individualdatasets . Inspired by recent non-maximum suppression-free 2D objects detection models, we propose a new architecture on point clouds . This approach aligns the outputs of the teachermodel and the student model in a permutation-invariant fashion, significantly simplifying knowledge distillation for the 3D detection task .…

Social engineering scams (SES) has been existed since the adoption of thetelecommunications by humankind . An earlier version of the scams includeleveraging premium phone service to charge the consumers and service providers . There are variety of techniques being considered to scam the people due to the advancements in digital data access capabilities andInternet technology .…

Many-to-one maps are ubiquitous in machine learning . In many problems it is useful to explore, understand, and sample from a model’s fibers . In this paper we show that popular generativearchitectures are ill-suited to such tasks . Motivated by this we introduce anovel generative architecture, a Bundle Network, based on the concept of afiber bundle from (differential) topology .…

Ego4D offers 3,025 hours of daily-life activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 855 uniquecamera wearers from 74 worldwide locations and 9 different countries . Portions of the video are accompanied by audio, audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event .…

We present a deep-learning artificial intelligence model that is capable oflearning and forecasting the late-inspiral, merger and ringdown of numericalrelativity waveforms that describe quasi-circular, spinning, non-precessingbinary black hole mergers . We harnessed theThetaGPU supercomputer at the Argonne Leadership Computing Facility to trainour AI model using a training set of 1.5 million waveforms .…

We study the convergence of a symmetrized Graph Laplacian matrix induced by a Gaussian kernel evaluated on pairs of embedded data, sampled from a manifold with boundary . We deduce theconvergence of the \emph{truncated Graph LaPlacian}, which is recentlynumerically observed in applications, and provide a detailed numerical investigation on simple manifolds .…

The essence of meaning conveyed by words is best described by a compass-like power-danger framework . Analysis of a disparate collection of large-scale English language corpora –literature, news, Wikipedia, talk radio, and social media — shows that naturallanguage exhibits a systematic bias toward safe, low danger words .…

Alibaba serves an indispensable role forhundreds of different travel scenarios from Fliggy, Taobao, Alipay apps, etc. To provide personalized recommendation service for users visiting differentscenarios, there are two critical issues to be carefully addressed . In this paper, we propose a novelScenario-Aware Ranking Network (SAR-Net) to address these issues .…

Development of new machine learning models is typically done on manuallycurated data sets, making them unsuitable for evaluating the models’ performance during operations . With this in mind, we developed a web-based visualization system that allows users to quickly gather headline performance numbers while maintaining confidence that the underlying data pipeline is functioning properly .…

The main process behind a software defect is that an error-prone scenariotriggers human error modes, which psychologists have observed to recur acrossdiverse activities . The proposed ideaemphasizes predicting the exact location and form of a possible defect . The approach was able to predict, atthe requirement phase, the exact locations and forms of 7 out of the 22 (31.8%) specific types of defects that were found in the code .…

Outfit recommendation often requires a complex and large model that incurs huge memory and time costs . The explosive number of outfit candidates amplifies the data sparsity problem, often leading to poor outfit representation . False NegativeDistillation (FND) exploits false-negative information from the teacher model while not requiring the ranking of all candidates .…

This paper presents a methodology to carry out compliance checking on reifiedI/O logic formulae . These are translated in SHACL (Shapes Constraint Language)shapes, a recent W3C recommendation to validate and reason with RDFtriplestores . Compliance checking is then enforced by validating RDF graphs describing states of affairs with respect to these ShACL shapes .…

Linear temporal logic (LTL) is a specification language for finite sequences called traces . It is widely used in program verification, motion planning inrobotics, process mining, and many other areas . Existing solutions suffer from two limitations: they do notscale beyond small formulas, and they may exhaust computational resources .…

In adversarial machine learning, new defenses against attacks on deeplearning systems are routinely broken soon after their release by more powerfulattacks . Forensic tools can offer a valuable complement to existing defenses, by tracing back a successful attack to its root cause .…

Interval computation widely used to certify computations that use floatingpoint operations to avoid pitfalls related to rounding error introduced byinaccurate operations . Despite its popularity and practical benefits, supportfor interval arithmetic is not standardized nor available in mainstamprogramming languages . We propose the first benchmark for intervalcomputations, coupled with reference solutions computed with exact arithmetic,and compare popular C and C++ libraries over different architectures, operatingsystems, and compilers .…

Training data for machine translation (MT) is often sourced from a multitude of large corpora that are multi-faceted in nature, e.g. containing contents from multiple domains or different levels of quality or complexity . In this work, we propose to optimize thisbalance jointly with MT model parameters to relieve system developers frommanual schedule design .…

The major finding, of this article, is an ensemble method that aims to solve the problem of finding the best candidate to represent the voters . We have convincing evidence that our algorithm is better than Instant-RunoffVoting, Preferential Block Voting, Single Transferable Vote, and First Past ThePost .…

POI recommendation is a key task in tourism information systems . Most conventional systems rank sightseeing spots based on popularity, reputation and similarities with users’ preferences . We propose amechanism to recommend POIs to tourists . Our mechanism include two components: one is a probabilistic model that reveals the user behaviors in tourism .…

We propose a novel conflict resolution framework for IoT services in multi-resident smart homes . The proposed framework employs a preferenceextraction model based on a temporal proximity strategy . We design a preferenceaggregation model using a matrix factorization-based approach . The concepts of current resident item matrix and idealresident item matrix are introduced as key criteria to cater to the conflictresolution framework .…

We present the meshfree Mixed Collocation Method (MCM) to solve themonodomain model for numerical simulation of cardiac electrophysiology . We apply MCM to simulate cardiac electrical propagation in 2D tissue sheets and 3D tissue slabs as well as in realistic large-scale biventricular anatomies .…

An interactive instruction following task has been proposed as a benchmark for learning to map natural language instructions and first-person vision intosequences of actions to interact with objects in a 3D simulated environment . We assume that this problem is due to the high sensitiveness of neural feature extraction to small changes invision and language inputs .…

In this paper, we study the statistical limits of deep learning techniques for solving elliptic partial differential equations . We focus on a prototype elliptic PDE: theSchrodinger equation on a hypercube with zero Dirichlet boundary condition . We establishupper and lower bounds for both methods, which improves upon concurrentlydeveloped upper bounds for this problem via a fast rate generalization bound .…

libdlr is a library implementing the recently introduced discreteLehmann representation (DLR) of imaginary time Green’s functions . DLR basisconsists of a collection of exponentials chosen by the interpolativedecomposition to ensure stable and efficient recovery of functions fromimaginary time or Matsbuara frequency samples .…

It is well known that an arbitrary $n$-qubit quantum state can be prepared with $O(n^2 2^{n/2)$straddling gates . In this work, we give an algorithm that fulfills the task . We also define a novelbinding complexity class, the “Schmidt decomposable” states, and give a circuit construction explanation for its unique property .…

We present a mathematical and numerical framework for thin-film fluid flow . We provide algorithmic details and an implementation of higher-order spatial and temporal discretisation of the underlying free boundary problem using the finite element method . We investigate the impact of the dynamic contact angle on the evolution of two benchmark problems:gravity-driven sliding droplets and the instability of a ridge .…

Text autoencoders are often used for unsupervised conditional text generation . We propose regularization techniques that facilitate learning meaningful operations in thelatent space . We adapt for a training scheme that learns to map an input bag to an output bag, including a novel loss function and neuralarchitecture .…

This paper is concerned with the time-dependent Maxwell’s equations for aplane interface between a negative material described by the Drude model and the vacuum . In a first paper, we have constructed a generalized Fourier transform which diagonalizesthe Hamiltonian that represents the propagation of transverse electric waves .…

In reinforcement learning, the environment dynamics is implicitly assumed to be stationary . This assumption of stationarity, while simplifying, can be unrealistic in many scenarios . In this work, we propose to examinethis continual reinforcement learning setting through the block contextual MDP(BC-MDP) framework .…

We consider the problem of black-box multi-objective optimization (MOO) usingexpensive function evaluations (also referred to as experiments) The goal is to approximate the true Pareto set of solutions by minimizing the total resource cost of experiments . We appropriately instantiate the principle of OSEsearch to derive efficient algorithms for the following four MOO problemsettings: 1) The most basic em single-fidelity setting, where experiments areexpensive and accurate; 2) Handling em black-boxes constraints { which cannot beevaluated without performing experiments; 3) The discrete multi-financesetting, where .…

We propose a new penalty for constructing models to recover an unknown signal from incomplete and inaccuratemeasurements . Mathematically, the springback penalty is a weakly convexfunction . It bears various theoretical and computational advantages of both the benchmark convex $ell_1 penalty and many of its non-convex surrogatesthat have been well studied in the literature .…

We describe a Question Answering (QA) dataset that contains complex questions with conditional answers . ConditionalQA is challenging for many of the existing QA models, especially in selecting answer conditions . We believe that this dataset willmotivate further research in answering complex questions over long documents .…

Elementary matrices play an important role in linear algebra applications . We show their properties on inverses and transpose using rewriting rules of ZX-calculus . We are able to depict any matrices of size 2^m\times 2^n by string diagrams without resorting to a diagrammatic normal form for matrices as shown in [Wang 2020].…

Recently, pretrained generativesequence-to-sequence (seq2seq) models have achieved great success in questionanswering . We propose a novel method for applying Transformer models to extractivequestion answering tasks . Viewing cross-attention as an architectural prior, we apply joint training to further improve QA performance .…