Transfer Learning for Human & AI

Existing position based point cloud filtering methods can hardly preservesharp geometric features . We propose a novel positionbased approach for feature-preserving point cloud filter . Unlike normalbased techniques, our method does not require the normal information . The coreidea is to first design a similarity metric to search the non-local similarpatches of a queried local patch .…

Existing pattern mining techniqueseither assume human dynamics is strictly periodic, or require the number of modes as input, or do not consider uncertainty in the sensor data . In this paper, we propose a novel clustering approach formodeling human behavior from time-series data .…

The sandwiched R\’enyi divergences of two finite-dimensional quantumstates play a distinguished role among the many quantum versions of R\’enidivergences . They are tight quantifiers of the trade-off between the two errorprobabilities in the strong converse domain of state discrimination . We show the same for the sandwiched divergence of two normalstates on an injective von Neumann algebra, thereby establishing theoperational significance of these quantities .…

Statistical shape modeling (SSM) characterizes anatomical variations in apopulation of shapes generated from medical images . SSM requires consistentshape representation across samples in shape cohort . Theseshape representations are then used to extract low-dimensional shapedescriptors that facilitate subsequent analyses in different applications .…

Deep NLP models have been shown to learn spurious correlations, leaving them brittle to input perturbations . We develop a Retrieve-Generate-Filter-Filter technique to create counterfactualevaluation and training data with minimal human supervision . Using anopen-domain QA framework and question generation model trained on original task data, we create counterfactsuals that are fluent, semantically diverse, andautomatically labeled.…

Automated audio captioning (AAC) is the task of automatically generating descriptions for general audio signals . We evaluate the performance of off-the-shelf models with aTransformer-based captioning approach . We show that sequences of audio embeddings can be processed using a Transformer encoder to produce higher-qualitycaptions .…

Sensitivity of deep-neural models to input noise is known to be a challenging problem . In NLP, model performance often deteriorates with naturally occurring noise, such as spelling errors . We present a thoroughevaluation of several state-of-the-art NLP systems’ robustness in multiple languages .…

Training dense passage representations via contrastive learning (CL) has been effective for Open-Domain Passage Retrieval (ODPR) We call such conflicts Contrastive Conflicts . We propose to solve it with a representation decoupling method, bydecoupling the passage representations into contextual sentence-level ones .…

In this article we propose an inverse analysis algorithm to find the best fit of multiple material parameters in different coupled multi-physics biofilmmodels . We use a nonlinear continuum mechanical approach to model biofilmdeformation that occurs in flow cell experiments .…

The Possibilistic Fuzzy Local Information C-Means (PFLICM) method is presented as a technique to segment side-look synthetic aperture sonar (SAS)imagery into distinct regions of the sea-floor . The chosen features and resulting segmentation from the image will be assessed based on a select quantitative clustering validitycriterion and a subset of the features that reach a desired threshold will be used for the segmentation process .…

Pre-trained language models (PLMs) have revolutionized the field of NLP, not only in the general domain but also in the biomedical domain . In this work we introduceeHealth, a biomedical PLM in Chinese built with a new pre-training framework . This new framework trains eHealth as a discriminator through both token-leveland sequence-level discrimination .…

Continual Learning (CL) is an emerging machine learning paradigm that aims to learn from a continuous stream of tasks without forgetting knowledge learned from previous tasks . To avoid performance decrease caused by forgetting, prior studies exploit episodic memory (EM), which stores a subset of the pastobserved samples while learning from new non-i.i.d.…

We introduce semantic form mid-tuning, an approach for transferring semanticknowledge from semantic meaning representations into transformer-based languageencoders . We learn to align the text of general sentences with structured semantic representations of those sentences . We show that this alignment can be learnedimplicitly via classification or directly via triplet loss .…

In this paper we propose a variant of New Q-Newton’s method Backtracking . The update rule of our method is $x\mapsto x-\gamma (x)w(x)$. Good theoretical guarantees are proven, in particular for systems ofpolynomial equations . In “generic situations”, we will also discuss a way to avoid that the limit of the constructed sequence is a solution of $H(x), but not of $F(x)=0$ The limit is a .…

Modern audio-visual datasets contain biases that undermine the real-world performance of models trained on that data . We introduce Spoken ObjectNet to remove some of these biases . This dataset expands upon ObjectNet, which is a bias-controlled image dataset . We detail our datacollection pipeline, which features several methods to improve caption quality, including automated language model checks .…

Given an embodied agent trained in anoiseless environment, our objective is to transfer the agent to a noisy environment where actuation and odometry sensor noise is present . We propose a light-weight, self-supervised adaptation for a visual navigationagent to generalize to unseen environment .…

P-Adapters are lightweight models that sit between the embedding layer and first attention layer of Large Language Models . They take LLM embeddings as input and output continuous prompts that are used to query the LLM . They showbetween 12-26% absolute improvement in precision and 36-50% absoluteimprovement in consistency over a baseline of only using natural languagequeries .…

Modern epidemiological analyses to understand and combat the spread of disease depend critically on access to, and use of, data . Tracing the provenance of such decisions back through open software to primary data would enhance the transparency of the decision-making process .…

Pun location is to identify the punning word (usually a word or phrase that makes the text ambiguous) in a given short text . Pun interpretation is to find out two different meanings of the word . DANN (Dual-Attentive Neural Network) is proposed for pun location, effectively integrates word senses and pronunciation with context information to address two kinds of pun at the same time .…

Any-to-any voice conversion technologies convert the vocal timbre of anutterance to any speaker even unseen during training . But in real-world scenarios, it is difficult to collect clean utterances of a speaker, and they are usually degraded by noises or reverberations .…

Many studies have shown that linguistic information is encoded inhidden word representations . Few have studied individual neurons to show how and in which neurons it is encoded . The common approach is to use an external probe to rank neurons according to their relevance to somelinguistic attribute, and to evaluate the obtained ranking using the same probethat produced it .…

Diminished reality is a technology that aims to remove objects from video images and fills in the missing region with plausible pixels . We propose an RGB-D image inpainting method using generative adversarialnetwork, which does not require multiple cameras . The experimental results verify the effectiveness of our proposed method, we propose late fusion approach that exploits the advantage of RGB and depth information each other to jointly restore texture and geometry of missing regions from a pair ofRGB and depth images .…

Imputation Balanced GAN (IB-GAN) is anovel method that joins data augmentation and classification in a one-stepprocess via an imputation-balancing approach . IB-GAN uses imputation andresampling techniques to generate higher quality samples from randomly maskedvectors than from white noise . It augments classification through aclass-balanced set of real and synthetic samples .…

Instead ofperturbing pixels, we use an encoder-decoder representation of the input image and perturb intermediate layers in the decoder . This changes the high-levelfeatures provided by the generative model . We formulatethis task as an optimization problem by minimizing the Wasserstein distancebetween the adversarial and initial images under a misclassificationconstraint .…

Two families of RL-based agents learn simultaneously to satisfy their objective . Each group learns a shared policy able to generalize andpolate over a wide range of behaviors . This is a step towards a fullyRL-based market simulator replicating complex market conditions, particularly suited to study the dynamics of the financial market under various scenarios .…

Single-image generative adversarial networks learn from the internaldistribution of a single training example to generate variations of it . SpecSinGAN is anunconditional generative architecture that takes a single one-shot sound effect(e.g., a footstep; a character jump) and produces novel variations .…

This paper proposes a regression market for wind agents to monetize datatraded among themselves for wind power forecasting . Existing literature on datamarkets often treats data disclosure as a binary choice or modulates the dataquality based on the mismatch between the offer and bid prices .…

Social concepts referring to non-physical objects are powerful tools to describe, index, and query the content of visual data . We propose the translation of recent theories about social concept representation into a software approach to represent them as multimodal frames, by integrating multisensory data .…

This paper proposes a novel active Simultaneous Localization and Mapping (SLAM) method with continuous trajectory optimization over a stochastic robot kinematic model . The proposed optimization method captures the coupling between localization and mapping in predicting uncertainty evolution and synthesizes highly informative sensing trajectories .…

Recent works have aimed toimprove optimal transport plans through the introduction of various forms of structure . We introduce novel order constraints into the optimal transportformulation to allow for structure . While there will are now quadratically many constraints as before, we prove a .roximatesolution…

We consider the solution of nonlinear equations in one real variable, the problem usually called by root finding . We argue that problems with just one variable are much simpler than problems with more variables . We provide animplementation of our ideas in C++, and make this code available under theMozilla Public License 2.0 .…

LAGr produces semantic parses by predicting node and edge labels for a complete multi-layer input-aligned graph . Thestrongly-supervised algorithm requires aligned graphs as inputs and infers alignments for originally unaligned target graphs using an approximate MAP inference procedure . On the COGS and CFQ compositionalgeneralization benchmarks the strongly- and weakly- supervised LAGR algorithmsachieve significant improvements upon the baseline seq2seq parsers.…

Modern software systems rely on machine learning models to make data-driven decisions based on interactions with users and systems . For broader adoption, this practice must accommodate software engineers without ML backgrounds, and (ii) providemechanisms to optimize for product goals .…

SGoLAM is ranked 2nd in the CVPR 2021 MultiON(Multi-Object Goal Navigation) challenge . It does not require any training of neural networks, it could be used in an off-the-shelfmanner, and amenable for fast generalization in new, unseen environments . The mapping module converts observations into an occupancy map, and the goal localization modulemarks the locations of goal objects .…

Some in industry and government now advocate a new technology to access targeted data: client-side scanning . CSS would enable on-device analysis of data in the clear . CSS by its nature createsserious security and privacy risks for all society while it can provide assistance for law enforcement is at best problematic, authors say .…

Style transfer is the task of rewriting an input sentence into a target style while approximately preserving its content . We find existing few-shot methods perform this task poorly, with a strong tendency to copy inputs verbatim . We report promising qualitative results for several attributetransfer directions, including sentiment transfer, text simplification, genderneutralization and text anonymization, all without retraining the model .…

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 .…

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 .…

This paper establishes non-asymptotic convergence of the cutoffs in Randomserial dictatorship in an environment with many students, many schools, andarbitrary student preferences . Convergence is shown to hold when the number ofschools, $m$ and number of students, $n$, satisfy the relation $m \ln m\ll n$, and we provide an example showing that this result is sharp .…

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 .…

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 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.…

Quantum many-body systems involving bosonic modes or gauge fields haveinfinite-dimensional local Hilbert spaces which must be truncated to perform simulations of real-time dynamics on classical or quantum computers . We show that if states in these models aretruncated by imposing an upper limit on each local quantum number,and if the initial state has low local quantum numbers, then an error at most$\epsilon$ can be achieved by choosing $Lambda$ to scale polylogarithmically with $E-Eigenstates .…

There is a growing interest in estimating the user agreement by considering the standard deviation of the scores, instead of only predicting the mean aestheticopinion score . With such loss, the model is encouraged to learn theuncertainty of the content that is relevant to the diversity of observers’opinions, i.e.,…

We study the classic online bipartite matching problem with a twist: offlinenodes are reusable any number of times . Every offline node $i$ becomesavailable $d$ steps after it was assigned to. We give the first approximation factor beating$0.5$ namely a $0.505$ approximation, by adapting and interpreting the powerful technique of Online Correlated Selection .…

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 .…

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 .…

Growth in technology has resulted in the large-scale collection and processing of Personally Identifiable Information by organizations that rundigital services such as websites . Several African countries have recently started enacting new dataprotection regulations due to recent technological innovations . Only 16 percent of third-party tracking companies that track users on a particular website are disclosed in the site’s privacy policy statements .…

We study housing markets as introduced by Shapley and Scarf (1974) We prove that the core of housingmarkets respects improvement in the following sense: given an allocation in thecore of $H$ where agent $a$ receives a house $h$ if the value of the house owned by $a$.…

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 .…