98,859 Hits in 6.1 sec

Domain-Specific Bias Filtering for Single Labeled Domain Generalization [article]

Junkun Yuan, Xu Ma, Defang Chen, Kun Kuang, Fei Wu, Lanfen Lin
2021 arXiv   pre-print
To tackle this challenging task, we propose a novel method called Domain-Specific Bias Filtering (DSBF), which initializes a discriminative model with the labeled source data and then filters out its domain-specific  ...  Domain generalization (DG) utilizes multiple labeled source datasets to train a generalizable model for unseen target domains.  ...  Index Terms-Domain generalization, single labeled multisource data, bias filtering, semantic feature projection. I.  ... 
arXiv:2110.00726v2 fatcat:p7sxwp2tkjgtlecjyy7zluoiaa

Visual Identification of Problematic Bias in Large Label Spaces [article]

Alex Bäuerle, Aybuke Gul Turker, Ken Burke, Osman Aka, Timo Ropinski, Christina Greer, Mani Varadarajan
2022 arXiv   pre-print
While visual analysis tools are of great help when investigating potential bias in DL models, none of the existing approaches have been designed for the specific tasks and challenges that arise in large  ...  Thus, domain experts need to be able to extract and reason about bias throughout models and datasets to make informed decisions.  ...  When talking to the domain experts, they also mentioned that they suspect bias patterns in the label-space, such as labels for certain professions being biased towards specific skin tones.  ... 
arXiv:2201.06386v1 fatcat:bcr3nfioqjeazlb3v4mvndutt4

fLPS 2.0: rapid annotation of compositionally-biased regions in biological sequences

Paul M. Harrison
2021 PeerJ  
In this version, the user is now able to restrict analysis to a specified subset of residue types, and also to filter for previously annotated domains to enable detection of discontinuous CB regions.  ...  In the output, protein CB regions are now labelled with bias classes reflecting the physico-chemical character of the biasing residues.  ...  Beyond the standard conception of DNA bias as either {GC} or {AT}, substantial tracts of other possible biases were observed, including strand-specific dearths of single bases (i.e., the bias classes {  ... 
doi:10.7717/peerj.12363 pmid:34760378 pmcid:PMC8557692 fatcat:etolmuykpzehdigc3tassgdynu

Online Domain Adaptation for Multi-Object Tracking

Adrien Gaidon, Eleonora Vig
2015 Procedings of the British Machine Vision Conference 2015  
Figure 1 : Online domain adaptation for MOT via Bayesian filtering coupled with multi-task adaptation of all detectors jointly.  ...  We integrate our domain adaptation strategy in a novel motion model combining learned deterministic models with standard Bayesian filtering (cf. figure above) inspired by the popular Bootstrap filter.  ...  Figure 1 : Online domain adaptation for MOT via Bayesian filtering coupled with multi-task adaptation of all detectors jointly.  ... 
doi:10.5244/c.29.3 dblp:conf/bmvc/GaidonV15 fatcat:dqgmoadprrae7ntgk6a4frldhi

Domain Generalization for Object Recognition with Multi-task Autoencoders [article]

Muhammad Ghifary and W. Bastiaan Kleijn and Mengjie Zhang and David Balduzzi
2015 arXiv   pre-print
We propose a new feature learning algorithm, Multi-Task Autoencoder (MTAE), that provides good generalization performance for cross-domain object recognition.  ...  We found that (denoising) MTAE outperforms alternative autoencoder-based models as well as the current state-of-the-art algorithms for domain generalization.  ...  [21] proposed a multi-task max-margin classifier, which we refer to as Undo-Bias, that explicitly encodes dataset-specific biases in feature space.  ... 
arXiv:1508.07680v1 fatcat:gubdug62jrekpneaxrghrnb66m

Domain Generalization for Object Recognition with Multi-task Autoencoders

Muhammad Ghifary, W. Bastiaan Kleijn, Mengjie Zhang, David Balduzzi
2015 2015 IEEE International Conference on Computer Vision (ICCV)  
We found that (denoising) MTAE outperforms alternative autoencoder-based models as well as the current state-of-the-art algorithms for domain generalization.  ...  We propose a new feature learning algorithm, Multi-Task Autoencoder (MTAE), that provides good generalization performance for crossdomain object recognition.  ...  The authors would like to thank Chen Fang for sharing the Unbiased Metric Learning code and useful discussions.  ... 
doi:10.1109/iccv.2015.293 dblp:conf/iccv/GhifaryKZB15 fatcat:wlobz6v5vzh5feo35kbedtczeq

Interpreting Expert Annotation Differences in Animal Behavior [article]

Megan Tjandrasuwita, Jennifer J. Sun, Ann Kennedy, Swarat Chaudhuri, Yisong Yue
2021 arXiv   pre-print
Our experiments on a dataset from behavioral neuroscience demonstrate that compared to baseline approaches, our method is more accurate at capturing annotator labels and learns interpretable temporal filters  ...  We propose a new method using program synthesis to help interpret annotation differences for behavior analysis.  ...  Programs with a single filter had slightly lower F1 scores compared to the disjunction. For the DTs, the single depth 1 DT is much simpler than 10 depth 5 DTs.  ... 
arXiv:2106.06114v1 fatcat:3b47gdldwja3zcg7m46g3e34nu

Encouraging Intra-Class Diversity Through a Reverse Contrastive Loss for Better Single-Source Domain Generalization [article]

Thomas Duboudin, Emmanuel Dellandréa, Corentin Abgrall, Gilles Hénaff, Liming Chen
2021 arXiv   pre-print
develop deep learning algorithms able to generalize from a single training domain where no information about the test domain is available at training time.  ...  Because data distributions can change dynamically in real-life applications once a learned model is deployed, in this paper we are interested in single-source domain generalization (SDG) which aims to  ...  Overcoming Biases The issue of missing patterns is related to both out-ofdistribution generalization and bias avoidance in deep networks.  ... 
arXiv:2106.07916v1 fatcat:l3uvtfa4sbecpprwoq4q5lgn6e

On the Benefits of Selectivity in Pseudo-Labeling for Unsupervised Multi-Source-Free Domain Adaptation [article]

Maohao Shen, Yuheng Bu, Gregory Wornell
2022 arXiv   pre-print
Existing methods for such multi-source-free domain adaptation typically train a target model using supervised techniques in conjunction with pseudo-labels for the target data, which are produced by the  ...  In particular, we develop an information-theoretic bound on the generalization error of the resulting target model that demonstrates an inherent bias-variance trade-off controlled by the subset choice.  ...  Selective-pseudo-labeling: The aforementioned issue can be fixed by only generating pseudo-labels for a specific subset of D t .  ... 
arXiv:2202.00796v2 fatcat:v2plt5t5gnbepfd2giywgugku4

Multi-Domain Spoken Language Understanding Using Domain- and Task-Aware Parameterization [article]

Libo Qin, Minheng Ni, Yue Zhang, Wanxiang Che, Yangming Li, Ting Liu
2021 arXiv   pre-print
We propose to improve the parameterization of this method by using domain-specific and task-specific model parameters to improve knowledge learning and transfer.  ...  Experiments on 5 domains show that our model is more effective for multi-domain SLU and obtain the best results.  ...  The first stage uses a filter to mask out those domain-general tokens which does not need domain-specific features.  ... 
arXiv:2004.14871v2 fatcat:npt5eb7zovbgthe7csqa7ki3gi

Natural Language Model Re-usability for Scaling to Different Domains

Young-Bum Kim, Alexandre Rochette, Ruhi Sarikaya
2016 Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing  
However, building new domains and tasks that need a separate set of models is a bottleneck for scaling to a large number of domains and experiences.  ...  The proposed technique uses a constrained decoding method with a universal slot tagging model sharing the same schema as the collection of slot taggers built for each domain.  ...  With Post-Filter, we simply provide the best hypothesis generated by the slot tagger that meets the domain schema constraints, by computing the full n-best of slots and filtering out the slot types that  ... 
doi:10.18653/v1/d16-1222 dblp:conf/emnlp/KimRS16 fatcat:cu2jzlmrqjeubgtniczsdt7fcm

Evaluation of Correctness in Unsupervised Many-to-Many Image Translation [article]

Dina Bashkirova, Ben Usman, Kate Saenko
2021 arXiv   pre-print
Current state-of-the art UMMI2I methods generate visually pleasing images, but, since for most pairs of real datasets we do not know which attributes are domain-specific and which are domain-invariant,  ...  target domain that preserves domain-invariant information of the input source image and inherits the domain-specific information from the guidance image.  ...  The overall idea is to choose two sets of attributes that vary in each domain and filter the original single-domain dataset accordingly.  ... 
arXiv:2103.15727v2 fatcat:fh2f3tmnzbf4rkgv5na5j6jn3u

Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus [article]

Jesse Dodge, Maarten Sap, Ana Marasović, William Agnew, Gabriel Ilharco, Dirk Groeneveld, Margaret Mitchell, Matt Gardner
2021 arXiv   pre-print
In this work we provide some of the first documentation for the Colossal Clean Crawled Corpus (C4; Raffel et al., 2020), a dataset created by applying a set of filters to a single snapshot of Common Crawl  ...  Then we explore the content of the text itself, and find machine-generated text (e.g., from machine translation systems) and evaluation examples from other benchmark NLP datasets.  ...  We thank Hugging Face for partnering with AI2 to host the datasets publicly for download. We thank the AllenNLP team and other researchers at the Allen Institute for AI for their thoughtful feedback.  ... 
arXiv:2104.08758v2 fatcat:s3gkabvc7bhf7f6kt4r6ff6t6q

Probing the Effect of Selection Bias on Generalization: A Thought Experiment [article]

John K. Tsotsos, Jun Luo
2022 arXiv   pre-print
But what are the limits of generalization given such bias, and up to what point might it be sufficient for a real problem task?  ...  One obvious way to deal with bias is to ensure a large enough training set, but this might be infeasible for many domains.  ...  Acknowledgements The authors wish to thank Ershad Banijamali, Iuliia Kotseruba, Amir Rasouli, Amir Rosenfeld, Brian Cantwell Smith, Sven Dickinson and Konstantine Tsotsos for their helpful comments and  ... 
arXiv:2105.09934v2 fatcat:yhyqayewzjdr7mj4whpevo5ypu

Automating Risk of Bias Assessment for Clinical Trials

Iain J Marshall, Joel Kuiper, Byron C Wallace
2015 IEEE journal of biomedical and health informatics  
in CDSR to produce a pseudo-annotated labeled corpus.  ...  burdensome for clinical researchers.  ...  risk of bias for a specific quality domain.  ... 
doi:10.1109/jbhi.2015.2431314 pmid:25966488 fatcat:7ke6fe47hzgmzdox6jd3hfgnmy
« Previous Showing results 1 — 15 out of 98,859 results