A key characteristic of work on deep learning and neural networks in general is that it relies on representations of the input that support generalization, robust inference, domain adaptation and other desirable functionalities. Much recent progress in the field has focused on efficient and effective methods for computing representations. In this paper, we propose an alternative method that is ...

We describe our prototyping of a functioning DLX microprocessor, based on the 32-bit instruction set architecture developed by Patterson and Hennessy. This architecture is an emerging academic standard, but to our knowledge has yet to be successfully prototyped. Our implementation of DLX is a 12" x 15" 2layer circuit board, containing 59 chips and running on a 2 MHz clock. Our machine was devel...

In this paper, we introduce the novel concept of densely connected layers into recurrent neural networks. We evaluate our proposed architecture on the Penn Treebank language modeling task. We show that we can obtain similar perplexity scores with six times fewer parameters compared to a standard stacked 2layer LSTM model trained with dropout (Zaremba et al., 2014). In contrast with the current ...

Punctuation prediction and disfluency prediction can improve downstream natural language processing tasks such as machine translation and information extraction. Combining the two tasks can potentially improve the efficiency of the overall pipeline system and reduce error propagation. In this work1, we compare various methods for combining punctuation prediction (PU) and disfluency prediction (...

Scalable video coding is important in a number of applications where video needs to be decoded and displayed at a variety of spatial and temporal resolution scales or when resilience to errors is necessary. In this paper, we investigate a temporaldomain approach for resolution scalable video coding. This approach was originally proposed to and since then has been included in the Moving Picture ...

In this thesis, we introduce two distinct approaches – distance-aware scheduling and 2layer discrete power control (DPC), which are able to enhance the transmission capacity of a large-scale wireless ad hoc network. Transmission capacity is an area spectrum efficiency metric of a wireless ad hoc network with an outage constraint. The motivation of studying these two approaches is to investigate...

ABR will be one of the primary available ATM service for carrying data. Since it is based on the use of excess bandwidth in the network, it is expected to have a lower cost usage. Although, it is intended for nonreal time applications, the inclusion of a minimum cell rate (MCR) makes it an attractive and economical candidate for the transmission of delay tolerant video applications. Therefore, ...

We deal with the problem of learning a general class of 2-layer neural networks in polynomial time. The considered neural networks consist of k linear threshold units on the hidden layer and an arbitrary binary output unit. We show NP-completeness of the consistency problem for classes that use an arbitrary set of binary output units containing a function which depends on all input dimensions. ...

Vector Quantization (VQ) and Subband Coding (SBC) are promising candidates for future generations of image coders : VQ is known to outperform scalar quantization (SQ) techniques, SBC overcomes the block structure problem of DCT transform coders and has better energy compaction properties. The paper compares the performances of two SBC-VQ schemes (a QMF-based SBC with trained-codebook VQ and a p...

The use of the Internet for entertainment continues to grow. The success of rich media web portals, such as: YouTube, Flickr, or Google Maps proves that people look for more and more interactive experience on the Web. In this paper, we briefly report our previous work related to the human-computer interaction discipline: we describe 2Layer Interface Paradigm (2LIP) a new way for designing inter...