Improving GAN with Neighbors Embedding and Gradient Matching

We propose two new techniques for training Generative Adversarial Networks (GANs) in the unsupervised setting. Our objectives are to alleviate mode collapse in GAN and improve the quality of the generated samples. First, we propose neighbor embedding, a manifold learning-based regularization to explicitly retain local structures of latent samples in the generated samples. This prevents generator from producing nearly identical data samples from different latent samples, and reduces mode collapse. We propose an inverse t-SNE regularizer to achieve this. Second, we propose a new technique, gradient matching, to align the distributions of the generated samples and the real samples. As it is challenging to work with high-dimensional sample distributions, we propose to align these distributions through the scalar discriminator scores. We constrain the difference between the discriminator scores of the real samples and generated ones. We further constrain the difference between the gradients of these discriminator scores. We derive these constraints from Taylor approximations of the discriminator function. We perform experiments to demonstrate that our proposed techniques are computationally simple and easy to be incorporated in existing systems. When Gradient matching and Neighbour embedding are applied together, our GN-GAN achieves outstanding results on 1D/2D synthetic, CIFAR-10 and STL-10 datasets, e.g. FID score of 30.80 for the STL-10 dataset. Our code is available at: https://github.com/tntrung/gan

A Rough Set Based Approach to Find Learners' Key Personality Attributes in an E-Learning Environment

One of the challenges in personalized e-learning research is how to find the unique learning strategies according to a learner’s personality characteristic. A learner’s personalitycharacteristic may have many attributes, and all of them may not have equal values. Correlation analysis, regression analysis, discriminator function, and educational psychology have been used to find solutions, but these methods have their shortcomings. This article proposes an improved approach based on rough set theory to find thekey personality attributes and evaluates the importance of these attributes. The approach has been successfully used in the actual e-learning environment for a major research university in China.

Lambda Calculus Characterizations of Poly-Time

We consider typed λ-calculi with pairing over the algebra W of words over {0, 1}, with a destructor and discriminator function. We show that the poly-time functions are precisely the functions (1) λ-representable using simple types, with abstract input (represented by Church-like terms) and concrete output (represented by algebra terms); (2) λ-representable using simple types, with abstract input and output, but with the input and output representations differing slightly; (3) λ-representable using polymorphic typing with type quantification ranging over multiplicative types only; (4) λ-representable using simple and list types (akin to ML style) with abstract input and output; and (5) λ-representable over the algebra of flat lists (in place of W), using simple types, with abstract input and output.