No abstract

The ability to build a model on a source task and subsequently adapt such model on a new target task is a pervasive need in many astronomical applications. The problem is generally known as transfer learning in machine learning, where domain adaptation is a popular scenario. An example is to build a predictive model on spectroscopic data to identify Supernovae IA, while subsequently trying to adapt such model on photometric data. In this paper we propose a new general approach to domain adaptation that does not rely on the proximity of source and target distributions. Instead we simply assume a strong similarity in model complexity across domains, and use active learning to mitigate the dependency on source examples. Our work leads to a new formulation for the likelihood as a function of empirical error using a theoretical learning bound; the result is a novel mapping from generalization error to a likelihood estimation. Results using two real astronomical problems, Supernova Ia classification and identification of Mars landforms, show two main advantages with our approach: increased accuracy performance and substantial savings in computational cost.

We address the problem of open-set authorship verification, a classification task that consists of attributing texts of unknown authorship to a given author when the testing set may differ significantly with the training set in terms of documents and candidate authors. We present an end-to-end model-building process that is universally applicable to a wide variety of corpora, with little to no modification or fine-tuning. It relies on transfer learning of a deep language model and uses a generative adversarial network and a number of text augmentation techniques to improve the model's generalization ability. The language model encodes documents of known and unknown authorship into a domain-invariant space, aligning document pairs as input to the classifier while keeping them separate. The resulting embeddings are used to train an ensemble of recurrent and quasi-recurrent neural networks. The entire pipeline is bidirectional; forward and backward pass results are averaged. We perform experiments on four traditional authorship verification datasets, a collection of machine learning papers mined from the web, and a large Amazon-Reviews dataset. Experimental results outperform baseline and state-of-the-art techniques, validating the proposed approach.

Device scaling engineering is facing major challenges in producing reliable transistors for future electronic technologies. With shrinking device sizes, the total circuit sensitivity to both permanent and transient faults has increased significantly. Research for fault tolerant processors has primarily focused on the conventional processor architectures. Neural network computing has been employed to solve a wide range of problems. This paper presents a design and implementation of a physical neural network that is resilient to permanent hardware faults. To achieve scalability, it uses tiled neuron clusters where neuron outputs are efficiently forwarded to the target neurons using source based spanning tree routing. To achieve fault resilience in the face of increasing number of permanent hardware failures, the design proactively preserves neural network computing performance by selectively replicating performance critical neurons. Furthermore, the paper presents a spanning tree recovery solution that mitigates disruption to distribution of neuron outputs caused by failed neuron clusters. The proposed neuron cluster design is implemented in Verilog. We studied the fault resilience performance of the described design using a RBM neural network trained for classifying handwritten digit images. Results demonstrate that our approach can achieve improved fault resilience performance by replicating only 5% most important neurons.

While enriching the user experiences, the development of mobile devices and applications introduces new security and privacy vulnerabilities for the remote services accessed by mobile device users. A trusted and usable authentication architecture for mobile devices is thus in high demand. In this paper, we leverage a unified structure, consisting of transparent TFTbased fingerprint sensors, touchscreen, and display, to propose a novel identity management mechanism that authenticates users of touch based mobile devices for accessing the local devices and remote services. Our solution differs from the previous onetime and enforced authentication approaches through two novel features: (i) user transparent authentication process, requiring neither password nor extra login steps and (ii) continuous identity management based on fingerprint biometric, where each user-to-device touch interaction is used toward authentication. Moreover, we introduce two different security scenarios, one for local identity management, and the second extended solution for remote identity management. Finally we employ TRUST (Trust Reinforcement based on Unified Structural Touch-display) to solve the identity challenge in cyber space.