Transfer Learning in a Transductive Setting

By Marcus Rohrbach, Sandra Ebert, Bernt Schiele. 2013

252 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (2017, background): “Apart from the inductive zero-shot learning set-up where the model has no access to neither visual nor side-information of unseen classes, transductive zero-shot learning approaches [28], [44], [45], [46], [47], [48], [49] use visual or semantic information of both seen and unseen classes without ha”
Learning to Self-Train for Semi-Supervised Few-Shot Classification (2019, background): “[25] proposed to further leverage external knowledge, such as the semantic attributes of categories, to solve not only few-shot but also zero-shot problems.”
Meta-Transfer Learning Through Hard Tasks (2019, methodology): “the word embedding models used for zero-shot learning [14], [55] and trained attribute models used for social relationship recognition [50].”
Feature Generating Networks for Zero-Shot Learning (2017, background): “On the other hand, instead of using only labeled data, [15, 38, 26] leverage unlabeled data from unseen classes in the transductive setting.”
Transductive Propagation Network for Few-shot Learning (2018, background): “Another category of transduction methods involves graph-based methods [30, 25, 27, 16, 3].”
Zero-Shot Learning with Generative Latent Prototype Model (2017, methodology): “ of virtual instances (NPC). Results are obtained using goog. nation of fc8, googand res fcfor Xand compare it with several state-of-the-art ZSL methods. They are SJE [31], HAP [38], ZSLwUA [39], PST [13], TMV [21], AMP [26], UDA [40] and UDICA [16]. For simple comparison, we use same settings and a”
Learning Transferrable Representations for Unsupervised Domain Adaptation (2016, methodology): “[24] utilizes a semi-supervised label propagation algorithm into the semisupervised transfer learning problem assuming access to few labeled examples and additional human specified semantic knowledge.”
KRISP: Integrating Implicit and Symbolic Knowledge for Open-Domain Knowledge-Based VQA (2020, background): “Some of the earliest versions of this work was relating to attributes [19, 67] or knowledge mined from the web [63], often for zero- or few-shot learning problems [20, 37, 62], as well as for finegrained classification [18].”
I Have Seen Enough: Transferring Parts Across Categories (2016, background): “This can be done by modeling semantic relatedness [38, 39], or by transferring other knowledge such as materials [5, 54], segments [45], parts [10, 49] or attributes [11, 29].”
Transductive Learning for Zero-Shot Object Detection (2019, background): “[27] explored the manifold structure of unseen classes by graph-based label propagation.”
Transductive Multi-view Embedding for Zero-Shot Recognition and Annotation (2014, methodology): “However, apart from suffering from the projection domain shift problem, [27] has limited ability to exploit multiple semantic representations/views.”
Learning to Propagate Labels: Transductive Propagation Network for Few-Shot Learning (2018, background): “Another category of transduction methods involves graph-based methods (Zhou et al., 2004; Wang and Zhang, 2006; Rohrbach et al., 2013; Fu et al., 2015).”
Billion-scale semi-supervised learning for image classification (2019, background): “Recent approaches [11, 35] revisited traditional label propagation algorithms (See [3, 9] for a review) to leverage the large YFCC unlabelled collection for low-shot learning.”
Zero-Shot Recognition via Semantic Embeddings and Knowledge Graphs (2018, background): “[40, 38] mined these associations from different internet sources.”
Incremental Learning-to-Learn with Statistical Guarantees (2018, background): “In particular, substantial improvement over “learning in isolation” (also known as independent task learning) is to be expected when the sample size per task is small, a setting which naturally arises in many applications [10, 30, 32, 37].”