Automatic Learning to Detect Concept Drift

Active Drift Detection with Meta learning (Meta-ADD) is a novel framework that learns to classify concept drift by tracking the changed pattern of error rates . Meta-ADD uses machine learning to learn to detect concept drifts and identify their types automatically, which candirectly support drift understand .…

Robustness Enhancement of Object Detection in Advanced Driver Assistance Systems ADAS

A unified system integrating a compact object detector and a surroundingenvironmental condition classifier for enhancing the robustness of objectdetection scheme in advanced driver assistance systems (ADAS) is proposed in this paper . The proposed system includes two main components: (1) a compactone-stage object detector which is expected to be able to perform at acomparable accuracy compared to state-of-the-art object detectors, and (2) an environmental condition detector that helps to send a warning signal to the cloud in case the self-driving car needs human actions due to the significance of the situation .…

Isolation schemes for problems on decomposable graphs

The Isolation Lemma is a self-reduction scheme that allows one to assume that a given instance of a problem has a unique solution . In this paper, we study a setting that is more typical for $\mathsf{NP}$-complete problems, and obtain partial derandomizations in theform of significantly decreasing the number of required random bits .…

Hallucination Improves Few Shot Object Detection

Learning to detect novel objects from few annotated examples is of greatpractical importance . A particularly challenging yet common regime occurs whenthere are extremely limited examples (less than three) One critical factor in improving few-shot detection is to address the lack of variation in training data .…

Leveraging Third Order Features in Skeleton Based Action Recognition

Recent skeleton-based actionrecognition methods extract features from 3D joint coordinates asspatial-temporal cues . We propose fusing third-orderfeatures in the form of angles into modern architectures, to robustly capture relationships between joints and body parts . This simple fusion with spatial-temporality graph neural networks achieves new state-of-the-artaccuracy in two large benchmarks, including NTU60 and NTU120, while employing fewer parameters and reduced run time .…

VQCPC GAN Variable length Adversarial Audio Synthesis using Vector Quantized Contrastive Predictive Coding

Generative Adversarial Networks are often adopted for the audio domain using fixed-size two-dimensionalspectrogram representations . VQCPC-GAN is an adversarial framework for synthesizing variable-length audio by exploiting Vector-Quantized Contrastive PredictiveCoding . The input noise z(characteristic in adversarial architectures) remains fixed over time, ensuring temporal consistency of global features of the generated content .…

Bring Your Own Codegen to Deep Learning Compiler

Deep neural networks (DNNs) have been ubiquitously applied in many applications . To achieve highmodel coverage with high performance, each accelerator vendor has to develop afull compiler stack to ingest, optimize, and execute the DNNs . To address these issues, this paper proposes an open source framework that enables users to only concentrate on the development of their own code generation tools by reusing as many as possible components inthe existing deep learning compilers .…

Orienting Point Clouds with Dipole Propagation

Establishing a consistent normal orientation for point clouds is anotoriously difficult problem in geometry processing . The normal direction of a point is a function of the local surface neighborhood; yet, point clouds do not disclosethe full underlying surface structure .…

Canonical Saliency Maps Decoding Deep Face Models

The Canonical Saliency Maps highlight facial features responsible for the decision made by a deep face model on a given image . Image-level maps highlight facialfeatures responsible for a DNN decision, thus helping to understand how a . DNN made a prediction on the image .…

Unsupervised Graph based Topic Modeling from Video Transcriptions

The model improvescoherence by exploiting neural word embeddings through a graph-based clusteringmethod . Unlike typical topic models, this approach works without knowing the true number of topics . Experimental results on the real-life multimodal dataset MuSe-CaR demonstrates that our approach extracts coherent and meaningfultopics, outperforming baseline methods .…

PreSizE Predicting Size in E Commerce using Transformers

PreSizE is a novel deeplearning framework which uses Transformers for accurate size prediction . It models the effect of both content-based attributes, such as brand andcategory, and the buyer’s purchase history on her size preferences . By encoding itemattributes, PreSZE better handles cold-start cases with unseen items, and cases where buyers have little past purchase data .…

Multipath Graph Convolutional Neural Networks

Graph convolution networks have recently garnered a lot of attention forrepresentation learning on non-Euclidean feature spaces . In this work, we propose a novel Multipath Graphconvolutional neural network that aggregates the output of multiple different shallow networks . Results show that the proposed method attains increased test accuracy but also requires fewer trainingepochs to converge .…

TimeGym Debugging for Time Series Modeling in Python

TimeGym Forecasting Debugging Toolkit is a Python library fortesting and debugging time series forecasting pipelines . It provides generic tests for forecasting pipelines fresh out of the box . The library enables forecasters to apply aTest-Driven Development approach to forecast modeling, using specified oraclesto generate artificial data with noise .…

ZEN 2 0 Continue Training and Adaption for N gram Enhanced Text Encoders

Pre-trained text encoders have drawn sustaining attention in natural languageprocessing (NLP) They have shown their capability in obtaining promising results indifferent tasks . We propose topre-train n-gram-enhanced Encoders with a large volume of data and advanced techniques for training . We try to extend the encoder to different languages as well as different domains, where it is confirmed that the samarchitecture is applicable to these varying circumstances and new state-of-the-art performance is observed from a long list of NLP tasks across the languages and domains .…

Distributive Justice and Fairness Metrics in Automated Decision making How Much Overlap Is There

The advent of powerful prediction algorithms led to increased automation of high-stake decisions regarding the allocation of scarce resources . This automation bears the risk of unwanted discrimination against vulnerable and historically disadvantaged groups . We argue that by cleanly distinguishing between prediction tasks and decision tasks, research on fair machine learning could take better advantage of the rich literature on distributive justice .…

Simplified Klinokinesis using Spiking Neural Networks for Resource Constrained Navigation on the Neuromorphic Processor Loihi

C. elegans shows chemotaxis using klinokinesis where the worm senses the concentration based on a single concentration sensor to compute the concentration gradient to perform foraging through gradient ascent/descenttowards the target concentration . The biomimeticimplementation requires complex neurons with multiple ion channel dynamics aswell as interneurons for control .…

HASCO Towards Agile HArdware and Software CO design for Tensor Computation

Tensor computations overwhelm traditional general-purpose computing devices . They callfor a holistic solution composed of both hardware acceleration and softwaremapping . Hardware/software (HW/SW) co-design optimizes the hardware andsoftware in concert and produces high-quality solutions . Hasco achieves a 1.25X to 1.44Xlatency reduction through HW/SW co-Design compared with developing the hardwareand software separately .…

Poisoning the Unlabeled Dataset of Semi Supervised Learning

Semi-supervised machine learning models learn from a (small) set of labeled training examples . We study a new class of vulnerabilities: poisoning attacks that modify theunlabeled dataset . By inserting maliciously-crafted unlabeled examples totaling just 0.1%of the dataset size, we can manipulate a model trained on this poisoned dataset .…

A Review on Oracle Issues in Machine Learning

Machine learning contrasts with traditional software development in that theoracle is the data, and the data is not always a correct representation of the problem that machine learning tries to model . We present a survey of the oracle issues found in machine learning and state-of-the-art solutions for dealing with these issues .…

Russian News Clustering and Headline Selection Shared Task

This paper presents the results of the Russian News Clustering and HeadlineSelection shared task . We propose tasks of Russian newsevent detection, headline selection, and headline generation . The presented datasets for eventdetection and headline selection are the first public Russian datasets for their tasks .…

On the limit of English conversational speech recognition

The study also considers the recently proposed conformer, and more advanced self-attention based language models . Their combination and decoding reaches a new record on Switchboard-300 and 10.0% WER on SWB and CHM parts of Hub5’00 with very simple LSTMmodels. Overall, the conformer showssimilar performance to the L STM; nevertheless, their combination and decode with an improved LM reaches new record .…

Modeling Social Readers Novel Tools for Addressing Reception from Online Book Reviews

Readers’ responses to literature have received scant attention to incomputational literary studies . The rise of social media offers an opportunity to capture a segment of these responses . Computationally modeling them allows oneto discover the overall non-professional discussion space about a work, including an aggregated summary of the work’s plot, an implicit ranking of theimportance of events, and the readers’ impressions of main characters .…

Goldilocks Just Right Tuning of BERT for Technology Assisted Review

Technology-assisted review (TAR) refers to iterative active learning workflows for document review in high recall retrieval (HRR) tasks . We find that the pre-trained BERT model reduces review volume by 30% in TAR workflows simulated on RCV1-v2 newswire collection . In contrast, linear models outperform BERT for simulated legal discovery topics on Jeb Bush e-mail collection .…

Weighted Least Squares Twin Support Vector Machine with Fuzzy Rough Set Theory for Imbalanced Data Classification

Support vector machines (SVMs) are powerful supervised learning tools developed to solve classification problems . However, SVMs are likely to performpoorly in the classification of imbalanced data . In this work, we propose an approach that efficiently used fuzzy rough set theory in weighted leastsquares twin support vector machine called FRLSTSVM for classification ofimbalanced data.…

Leveraging Deep Representations of Radiology Reports in Survival Analysis for Predicting Heart Failure Patient Mortality

Utilizing clinical texts in survival analysis is difficult because they are largely unstructured . Current automatic extraction models fail to capture textual information comprehensively . They typically require a large amount of data and high-quality annotations for training . In this work, we present a novel method ofusing BERT-based representations of clinical texts as covariates for proportional hazards models to predict patient survival outcomes .…

Automated Estimation of Total Lung Volume using Chest Radiographs and Deep Learning

Total lung volume is an important quantitative biomarker and is used for the assessment of restrictive lung diseases . In this study, we investigate theperformance of several deep-learning approaches for automated measurement of total lung volume from chest radiographs . We demonstrate, for the firsttime, that state-of-the-art deep learning solutions can accurately measuretotal lung volume .…

Impact of Gender Debiased Word Embeddings in Language Modeling

Gender, race and social biases have recently been detected as evident examples of unfairness in applications of Natural Language Processing . A keypath towards fairness is to understand, analyse and interpret our data and algorithms . Recent studies have shown that the human-generated data used intraining is an apparent factor of getting biases .…