Estimating labels from label proportions

By Novi Quadrianto, Alex Smola, Tibério S. Caetano, Quoc V. Le. 2008

276 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.

In Artificial Intelligence

Citations in the corpus

Fast Learning from Distributed Datasets without Entity Matching (2016, background): “Alth ough, it is not as restrictive as may be expected at first: if just one peer has labels, then all can obtain labels on their own data, via le rning from label proportions[23, 24]: the label handling peer computes the label proport ions per each block; the “bags” are defined by examples matching”
On classification with bags, groups and sets (2014, background): “Other scenarios where only training objects are bags are learning about individuals from group statistics [28], aggregate output learning [41] and learning with label proportions [43], independent names for very related ideas.”
Rademacher Observations, Private Data, and Boosting (2015, background): “More generally, the coordinates of some vector z ∈ Rd are denoted z1, z2, ..., zd.”
Hilbert space embeddings of conditional distributions with applications to dynamical systems (2009, methodology): “A special case of this conditional embedding was employed in Quadrianto et al. (2008) for discrete conditioning variables x (multiclass labels), where a delta kernel is applied onx and the space of observations (y) is partitioned according to the value of x. In this case, the conditional embedding”
Who Supported Obama in 2012?: Ecological Inference through Distribution Regression (2015, background): “A related line of work, termed“learning from label propor-tions” by some authors [24, 15, 30, 21], has the individual-level goal in mind, and aims to build a classiﬁer for individual instances based only on group level label proportions.”
(Almost) No Label No Cry (2014, methodology): “In this paper, we have shown that efficient learning in the LLP setting is possible, for general loss functions, via the mean operator and without resorting to the homogeneity assumption.”
Generalized Expectation Criteria for Semi-Supervised Learning with Weakly Labeled Data (2010, background): “Similar to label regularization, Quadrianto et al. (2008) uses label proporti ns to learn classifiers, though there are some interesting differences from our work.”
Learning from measurements in exponential families (2009, background): “In Quadrianto et al. (2008), examples are partitioned into sets, and all (x, y) 7→ I[f(x) = a, y = b] are measured, where f(x) is the set to which x belongs.”
Classification without labels: learning from mixed samples in high energy physics (2017, background): “[34] to distinguish jets originating from quarks from those originating from gluons using only class proportions during training; this paradigm is known as learning from label proportions (LLP) [35, 36].”
Exposed! A Survey of Attacks on Private Data (2017, background): “Sankararaman et al. (2009) presented a formal model of tracing attacks based on hypothesis testing.”
$\propto$SVM for learning with label proportions (2013, methodology): “Compared to (Rüeping, 2010; Quadrianto et al., 2009), ∝SVM requires no restrictive assumptions on the data.”
Degrees of supervision (2011, background): “• Label proportions: Here the supervision is given in the form of prior class probabilities Pi over bags of points S1, . . . , SM (Quadrianto et al., 2009).”
Unsupervised Supervised Learning II: Margin-Based Classification without Labels (2011, background): “The work in [10] estimates labels given several datasets with different (known) label proportions.”
From Group to Individual Labels Using Deep Features (2015, background): “Given this information, the goal is to predict the label of each instance within the bags [15, 27, 33].”
Loss factorization, weakly supervised learning and label noise robustness (2016, methodology): “The name mean operator, or mean map, is borrowed from the theory of Hilbert space embedding [Quadrianto et al., 2009](1).”
Weakly supervised learning via statistical sufficiency (2016, methodology): “The name mean operator, or mean map, is borrowed from the theory of Hilbert space embedding [Quadrianto et al., 2009].”