Deep networks have been widely used in many domains in recent years. However, the pre-training of deep networks is time consuming with greedy layer-wise algorithm, and the scalability of this algorithm is greatly restricted by its inherently sequential nature where only one hidden layer can be trained at one time. In order to speed up the training of deep networks, this paper mainly focuses on ...

Kernel machines are a powerful class of methods for classification and regression. Making kernel fast scalable to large data, however, is still challenging problem due the need storing operating on Gram matrix. In this paper we propose novel approach sample condensation machines, preferably without impairing performance. To our best knowledge, there no previous work with same goal reported in l...

How do we perform efficient inference while retaining high translation quality? Existing neural machine models, such as Transformer, achieve performance, but they decode words one by one, which is inefficient. Recent non-autoregressive models speed up the inference, their quality still inferior. In this work, propose DSLP, a highly and high-performance model for translation. The key insight to ...

Dynamic voltage frequency scaling (DVFS) has been widely adopted for run-time power management of various processing units. In the case neural units (NPUs), network applications is required to adjust and every layer consider behavior performance each layer. Unfortunately, DVFS inappropriate layer-wise NPUs due long latency compared with execution time. Because fast enough keep up layer, we prop...

A switching network of depth d is a layered graph with d layers and n vertices in each layer. The edges of the switching network do not cross between layers and in each layer the edges form a partial matching. A switching network defines a stochastic process over Sn that starts with the identity permutation and goes through the layers of the network from first to last, where for each layer and ...