Mahdi Ghorbani

Mahdi Ghorbani, Emilien Bauer, Tobias Grosser et al.

Tensor algebra is a crucial component for data-intensive workloads such as machine learning and scientific computing. As the complexity of data grows, scientists often encounter a dilemma between the highly specialized dense tensor algebra and efficient structure-aware algorithms provided by sparse tensor algebra. In this paper, we introduce DASTAC, a framework to propagate the tensors's captured high-level structure down to low-level code generation by incorporating techniques such as automatic data layout compression, polyhedral analysis, and affine code generation. Our methodology reduces memory footprint by automatically detecting the best data layout, heavily benefits from polyhedral optimizations, leverages further optimizations, and enables parallelization through MLIR. Through extensive experimentation, we show that DASTAC achieves 1 to 2 orders of magnitude speedup over TACO, a state-of-the-art sparse tensor compiler, and StructTensor, a state-of-the-art structured tensor algebra compiler, with a significantly lower memory footprint.

Amir Shaikhha, Marios Kelepeshis, Mahdi Ghorbani

We introduce a framework for automatically choosing data structures to support efficient computation of analytical workloads. Our contributions are twofold. First, we introduce a novel low-level intermediate language that can express the algorithms behind various query processing paradigms such as classical joins, groupjoin, and in-database machine learning engines. This language is designed around the notion of dictionaries, and allows for a more fine-grained choice of its low-level implementation. Second, the cost model for alternative implementations is automatically inferred by combining machine learning and program reasoning. The dictionary cost model is learned using a regression model trained over the profiling dataset of dictionary operations on a given hardware architecture. The program cost model is inferred using static program analysis. Our experimental results show the effectiveness of the trained cost model on micro benchmarks. Furthermore, we show that the performance of the code generated by our framework either outperforms or is on par with the state-of-the-art analytical query engines and a recent in-database machine learning framework.

Mahdi Ghorbani, Fahimeh Fooladgar, Shohreh Kasaei

Deep neural network architectures have attained remarkable improvements in scene understanding tasks. Utilizing an efficient model is one of the most important constraints for limited-resource devices. Recently, several compression methods have been proposed to diminish the heavy computational burden and memory consumption. Among them, the pruning and quantizing methods exhibit a critical drop in performances by compressing the model parameters. While the knowledge distillation methods improve the performance of compact models by focusing on training lightweight networks with the supervision of cumbersome networks. In the proposed method, the knowledge distillation has been performed within the network by constructing multiple branches over the primary stream of the model, known as the self-distillation method. Therefore, the ensemble of sub-neural network models has been proposed to transfer the knowledge among themselves with the knowledge distillation policies as well as an adversarial learning strategy. Hence, The proposed ensemble of sub-models is trained against a discriminator model adversarially. Besides, their knowledge is transferred within the ensemble by four different loss functions. The proposed method has been devoted to both lightweight image classification and encoder-decoder architectures to boost the performance of small and compact models without incurring extra computational overhead at the inference process. Extensive experimental results on the main challenging datasets show that the proposed network outperforms the primary model in terms of accuracy at the same number of parameters and computational cost. The obtained results show that the proposed model has achieved significant improvement over earlier ideas of self-distillation methods. The effectiveness of the proposed models has also been illustrated in the encoder-decoder model.