One of the core aspects of human-human interaction is the ability to recognize and respond to the emotional and cognitive states of the other person, leaving human-computer interaction systems, at their core, to perfo...
详细信息
ISBN:
(纸本)9783319586281;9783319586274
One of the core aspects of human-human interaction is the ability to recognize and respond to the emotional and cognitive states of the other person, leaving human-computer interaction systems, at their core, to perform many of the same tasks.
Estimating heterogeneous treatment effects (HTE) from observational studies is rising in importance due to the widespread accumulation of data in many fields. Due to the selection bias behind the inaccessibility of co...
详细信息
Estimating heterogeneous treatment effects (HTE) from observational studies is rising in importance due to the widespread accumulation of data in many fields. Due to the selection bias behind the inaccessibility of counterfactual data, the problem differs fundamentally from supervised learning in a challenging way. However, existing works on modeling selection bias and corresponding algorithms do not naturally generalize to non-binary treatment spaces. To address this limitation, we propose to use mutual information to describe selection bias in estimating HTE and derive a novel error bound using the mutual information between the covariates and the treatments, which is the first error bound to cover general treatment schemes including multinoulli or continuous spaces. We then bring forth theoretically justified algorithms, the Mutual Information Treatment Network (MitNet), using adversarial optimization to reduce selection bias and obtain more accurate HTE estimations. Our algorithm reaches remarkable performance in both simulation study and empirical evaluation.
The support/query episodic training strategy has been widely applied in modern meta learning algorithms. Supposing the n training episodes and the test episodes are sampled independently from the same environment, pre...
详细信息
ISBN:
(纸本)9781713871088
The support/query episodic training strategy has been widely applied in modern meta learning algorithms. Supposing the n training episodes and the test episodes are sampled independently from the same environment, previous work has derived a generalization bound of O(1/root n) for smooth non-convex functions via algorithmic stability analysis. In this paper, we provide fine-grained analysis of stability and generalization for modern meta learning algorithms by considering more general situations. Firstly, we develop matching lower and upper stability bounds for meta learning algorithms with two types of loss functions: (1) nonsmooth convex functions with alpha-Holder continuous subgradients (alpha is an element of[0, 1));(2) smooth (including convex and non-convex) functions. Our tight stability bounds show that, in the nonsmooth convex case, meta learning algorithms can be inherently less stable than in the smooth convex case. For the smooth non-convex functions, our stability bound is sharper than the existing one, especially in the setting where the number of iterations is larger than the number n of training episodes. Secondly, we derive improved generalization bounds for meta learning algorithms that hold with high probability. Specifically, we first demonstrate that, under the independent episode environment assumption, the generalization bound of O(1/root n) via algorithmic stability analysis is near optimal. To attain faster convergence rate, we show how to yield a deformed generalization bound of O(ln n/n) with the curvature condition of loss functions. Finally, we obtain a generalization bound for meta learning with dependent episodes whose dependency relation is characterized by a graph. Experiments on regression problems are conducted to verify our theoretical results.
This paper studies kernel regression problems. The focus is on studying kernel algorithms that use the least squares criterion and developing methods so that the solution in the dual observation space intelligently ch...
详细信息
ISBN:
(纸本)0780383591
This paper studies kernel regression problems. The focus is on studying kernel algorithms that use the least squares criterion and developing methods so that the solution in the dual observation space intelligently chooses training examples. The Least Squares - Support Vector Machine (LS-SVM) and variants have attracted researchers as the solution to nonlinear problems can be formulated as an optimization problem that involves finding a solution to a set of linear equations in the primal or dual spaces. A drawback of using the LS-SVM is that the solution is not sparse, but involves a solution to a set of linear equations in the dual space that is dependent on the number of observations. This paper discusses an on-line algorithm that selectively chooses to add and delete training observations. Through examples we show that this algorithm can outperform LS-SVM solutions that use a larger window of randomly trained examples.
There are several learning methods which are suitable for neural networks. In this paper two of them are described Back-propagation (BP) and Genetic (GA) algorithms. These learning methods are compared here and they a...
详细信息
ISBN:
(纸本)9781424419456
There are several learning methods which are suitable for neural networks. In this paper two of them are described Back-propagation (BP) and Genetic (GA) algorithms. These learning methods are compared here and they are used for the control of modem telecommunication network nodes.
A class of unsupervised algorithms known as competitive learning (CL) was investigated for its application as an adaptive control mechanism for an educational toy. Two variants of CL were used, hard competitive learni...
详细信息
ISBN:
(纸本)0852966903
A class of unsupervised algorithms known as competitive learning (CL) was investigated for its application as an adaptive control mechanism for an educational toy. Two variants of CL were used, hard competitive learning (HCL) and soft competitive learning (SCL). It was clearly shown that CL was suitable for the unsupervised clustering needed in an autonomous robotic toy. SCL was found to out-perform HCL in the more challenging test cases examined. Furthermore, simulations indicated that radial basis functions may be used within the constraints of the hardware system if the exponential function was replaced with a lookup table equivalent of a least 15 elements.
This paper wants to supplement computational tests of deep learning vision algorithms with a sociologically grounded performance test of three widely used vision algorithms on Facebook images (Clarifai, Google Vision ...
详细信息
ISBN:
(纸本)9780998133102
This paper wants to supplement computational tests of deep learning vision algorithms with a sociologically grounded performance test of three widely used vision algorithms on Facebook images (Clarifai, Google Vision and Inception-v3). The test shows poor results and the paper suggests the use of a two-level labeling model that combines features with theoretically inspired accounts of the social value of pictures for uploaders. The paper contributes a suggestion for labeling categories that connects the two levels, and in conclusion discusses both advantages and disadvantages in accelerating user profiling through a better understanding of the incentives to upload images in the data-driven algorithmic society.
A RCF design methodology for the minimization of inter-MG coupling coefficient is proposed based on machine-learning algorithms. Lower coupling coefficient can be realized by the proposed methodology, compared with th...
详细信息
ISBN:
(纸本)9781943580705
A RCF design methodology for the minimization of inter-MG coupling coefficient is proposed based on machine-learning algorithms. Lower coupling coefficient can be realized by the proposed methodology, compared with that of the previously reported RCF.
We study the problem of (learning) algorithm comparison, where the goal is to find differences between models trained with two different learning algorithms. We begin by formalizing this goal as one of finding disting...
详细信息
We study the problem of (learning) algorithm comparison, where the goal is to find differences between models trained with two different learning algorithms. We begin by formalizing this goal as one of finding distinguishing feature transformations, i.e., input transformations that change the predictions of models trained with one learning algorithm but not the other. We then present MODELDIFF, a framework that leverages data-models (Ilyas et al., 2022) to compare learning algorithms based on how they use training data. We demonstrate MODELDIFF through three case studies, comparing models trained with/without data augmentation, with/without pre-training, and with different SGD hyperparameters. Our code is available at https://***/MadryLab/modeldiff.
Understanding large data sets is one of the most important and challenging problem in modern days. Exploration of genetic data sets composed of high-dimensional feature vectors can be treated as a leading example in t...
详细信息
暂无评论