Co-Training and Expansion: Towards Bridging Theory and Practice

By Maria-Florina Balcan, Avrim Blum, Ke Yang. Neural Information Processing Systems, 2004

333 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/0eaa981275145025c5064107c9ec7e58fec39f6b

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

Multi-view learning overview: Recent progress and new challenges (2017, background): “Thus several weaker assumptions have been considered [40] [41].”
Multi-view clustering: A survey (2018, unknown): “Thus, several weaker assumptions such as the weak conditional dependence assumption[20], much weaker “expansion” assumption[21], and the difference assumption[22] have been investigated.”
Methodologies for Cross-Domain Data Fusion: An Overview (2015, background): “Consequently, several weaker alternatives [6] have thus been considered.”
Interacting meaningfully with machine learning systems: Three experiments (2009, background): “The co-training (Blum and Mitchell 1998; Kiritchenko and Matwin 2001; Balcan et al. 2005) approach to semi-supervised learning employs two classifiers that work on the same data but have two different "views" of the data through independent sets of features.”
Co-Training for Domain Adaptation (2011, methodology): “Recent work [2] weakens this requirement significantly to a condition of -expandability.”
Hyperparameter Learning for Graph Based Semi-supervised Learning Algorithms (2006, background): “(Balcan et al., 2004) proposed so-called expansion condition to relax the conditional independence assumption.”
Co-regularized Alignment for Unsupervised Domain Adaptation (2018, methodology): “[8] used the idea of co-training for semi-supervised domain adaptation (assuming a few target labeled examples are available) by finding a suitable split of the features into two sets based on the notion of -expandibility [1].”
Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss (2021, background): “Co-training and multi-view learning are related settings which leverage two distinct “views” (i.e., feature subsets) of the data (Balcan et al., 2005, Blum and Mitchell, 1998, Dasgupta et al., 2002).”
A survey on semi-supervised learning (2019, background): “They formulated a single optimization problem closely related to co-training, incorporating both the requirement that the feature sets should be disjoint and the expansion property of Balcan et al. (2005). They showed promising results for this approach on a partially synthetic data set, where multi”
Leveraging Just a Few Keywords for Fine-Grained Aspect Detection Through Weakly Supervised Co-Training (2019, background): “Several subsequent works have relaxed this assumption and shown co-training to be effective in much more general settings (Balcan et al., 2005; Chen et al., 2011; Collins and Singer, 1999; Clark et al., 2018).”
Semi-supervised On-Line Boosting for Robust Tracking (2008, background): “4 There are also relations to the co-training [25] assumptions, i.”
Self-supervised Learning from a Multi-view Perspective (2020, background): “Prior work (Balcan et al., 2005; Du et al., 2010) have compared these two assumptions and pointed out the former one ( I ( X ; S | T ) ≈ 0 ) is too strong and not likely to hold in practice.”
Introduction to Semi-Supervised Learning (2009, background): “Further theoretical analysis of Co-Training can be found in [12, 10, 53].”
Bilingual Co-Training for Sentiment Classification of Chinese Product Reviews (2011, background): “Balcan, Blum, and Yang (2005) prove that a weaker assumption called ε -expansion is sufﬁcient for iterative co-training to succeed.”
A survey of multi-view machine learning (2013, background): “It was shown that the second assumption of co-training can be relaxed to a weaker expansion assumption on the underlying data distribution for iterative co-training to succeed, given appropriately strong PAC-learning algorithms on each view, and the expansion assumption is to some extent necessary a”