Ehsan Saboori

MohammadHossein AskariHemmat, Reyhane Askari Hemmat, Alex Hoffman et al.

In this paper we study the effects of quantization in DNN training. We hypothesize that weight quantization is a form of regularization and the amount of regularization is correlated with the quantization level (precision). We confirm our hypothesis by providing analytical study and empirical results. By modeling weight quantization as a form of additive noise to weights, we explore how this noise propagates through the network at training time. We then show that the magnitude of this noise is correlated with the level of quantization. To confirm our analytical study, we performed an extensive list of experiments summarized in this paper in which we show that the regularization effects of quantization can be seen in various vision tasks and models, over various datasets. Based on our study, we propose that 8-bit quantization provides a reliable form of regularization in different vision tasks and models.

Saad Ashfaq, MohammadHossein AskariHemmat, Sudhakar Sah et al.

Deep Learning has been one of the most disruptive technological advancements in recent times. The high performance of deep learning models comes at the expense of high computational, storage and power requirements. Sensing the immediate need for accelerating and compressing these models to improve on-device performance, we introduce Deeplite Neutrino for production-ready optimization of the models and Deeplite Runtime for deployment of ultra-low bit quantized models on Arm-based platforms. We implement low-level quantization kernels for Armv7 and Armv8 architectures enabling deployment on the vast array of 32-bit and 64-bit Arm-based devices. With efficient implementations using vectorization, parallelization, and tiling, we realize speedups of up to 2x and 2.2x compared to TensorFlow Lite with XNNPACK backend on classification and detection models, respectively. We also achieve significant speedups of up to 5x and 3.2x compared to ONNX Runtime for classification and detection models, respectively.

Darshan C. Ganji, Saad Ashfaq, Ehsan Saboori et al.

A lot of recent progress has been made in ultra low-bit quantization, promising significant improvements in latency, memory footprint and energy consumption on edge devices. Quantization methods such as Learned Step Size Quantization can achieve model accuracy that is comparable to full-precision floating-point baselines even with sub-byte quantization. However, it is extremely challenging to deploy these ultra low-bit quantized models on mainstream CPU devices because commodity SIMD (Single Instruction, Multiple Data) hardware typically supports no less than 8-bit precision. To overcome this limitation, we propose DeepGEMM, a lookup table based approach for the execution of ultra low-precision convolutional neural networks on SIMD hardware. The proposed method precomputes all possible products of weights and activations, stores them in a lookup table, and efficiently accesses them at inference time to avoid costly multiply-accumulate operations. Our 2-bit implementation outperforms corresponding 8-bit integer kernels in the QNNPACK framework by up to 1.74x on x86 platforms.

Sudhakar Sah, Darshan C. Ganji, Matteo Grimaldi et al.

We introduce MCUBench, a benchmark featuring over 100 YOLO-based object detection models evaluated on the VOC dataset across seven different MCUs. This benchmark provides detailed data on average precision, latency, RAM, and Flash usage for various input resolutions and YOLO-based one-stage detectors. By conducting a controlled comparison with a fixed training pipeline, we collect comprehensive performance metrics. Our Pareto-optimal analysis shows that integrating modern detection heads and training techniques allows various YOLO architectures, including legacy models like YOLOv3, to achieve a highly efficient tradeoff between mean Average Precision (mAP) and latency. MCUBench serves as a valuable tool for benchmarking the MCU performance of contemporary object detectors and aids in model selection based on specific constraints.

Saad Ashfaq, Alexander Hoffman, Saptarshi Mitra et al.

The proliferation of edge devices has unlocked unprecedented opportunities for deep learning model deployment in computer vision applications. However, these complex models require considerable power, memory and compute resources that are typically not available on edge platforms. Ultra low-bit quantization presents an attractive solution to this problem by scaling down the model weights and activations from 32-bit to less than 8-bit. We implement highly optimized ultra low-bit convolution operators for ARM-based targets that outperform existing methods by up to 4.34x. Our operator is implemented within Deeplite Runtime (DeepliteRT), an end-to-end solution for the compilation, tuning, and inference of ultra low-bit models on ARM devices. Compiler passes in DeepliteRT automatically convert a fake-quantized model in full precision to a compact ultra low-bit representation, easing the process of quantized model deployment on commodity hardware. We analyze the performance of DeepliteRT on classification and detection models against optimized 32-bit floating-point, 8-bit integer, and 2-bit baselines, achieving significant speedups of up to 2.20x, 2.33x and 2.17x, respectively.

MohammadHossein AskariHemmat, Ahmadreza Jeddi, Reyhane Askari Hemmat et al.

Quantization lowers memory usage, computational requirements, and latency by utilizing fewer bits to represent model weights and activations. In this work, we investigate the generalization properties of quantized neural networks, a characteristic that has received little attention despite its implications on model performance. In particular, first, we develop a theoretical model for quantization in neural networks and demonstrate how quantization functions as a form of regularization. Second, motivated by recent work connecting the sharpness of the loss landscape and generalization, we derive an approximate bound for the generalization of quantized models conditioned on the amount of quantization noise. We then validate our hypothesis by experimenting with over 2000 models trained on CIFAR-10, CIFAR-100, and ImageNet datasets on convolutional and transformer-based models.

Sudhakar Sah, Ravish Kumar, Darshan C. Ganji et al.

Activation functions introduce non-linearity into Neural Networks, enabling them to learn complex patterns. Different activation functions vary in speed and accuracy, ranging from faster but less accurate options like ReLU to slower but more accurate functions like SiLU or SELU. Typically, same activation function is used throughout an entire model architecture. In this paper, we conduct a comprehensive study on the effects of using mixed activation functions in YOLO-based models, evaluating their impact on latency, memory usage, and accuracy across CPU, NPU, and GPU edge devices. We also propose a novel approach that leverages Neural Architecture Search (NAS) to design YOLO models with optimized mixed activation functions.The best model generated through this method demonstrates a slight improvement in mean Average Precision (mAP) compared to baseline model (SiLU), while it is 22.28% faster and consumes 64.15% less memory on the reference NPU device.

Sudhakar Sah, Ravish Kumar, Honnesh Rohmetra et al.

High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT) model, a pre-processing block to object detection models like DETR/YOLOS aimed to reduce runtime memory and increase throughput by using a novel approach to identify background tokens in the image. The background tokens can be pruned completely or partially before feeding to a ViT based object detector. We use the semantic information provided by segmentation map and/or bounding box annotation to train a few layers of ViT to classify tokens to either foreground or background. Using 2 layers and 10 layers of BAViT, background and foreground tokens can be separated with 75% and 88% accuracy on VOC dataset and 71% and 80% accuracy on COCO dataset respectively. We show a 2 layer BAViT-small model as pre-processor to YOLOS can increase the throughput by 30% - 40% with a mAP drop of 3% without any sparse fine-tuning and 2% with sparse fine-tuning. Our approach is specifically targeted for Edge AI use cases.

Martin Ferianc, Anush Sankaran, Olivier Mastropietro et al.

Neural networks (NNs) are making a large impact both on research and industry. Nevertheless, as NNs' accuracy increases, it is followed by an expansion in their size, required number of compute operations and energy consumption. Increase in resource consumption results in NNs' reduced adoption rate and real-world deployment impracticality. Therefore, NNs need to be compressed to make them available to a wider audience and at the same time decrease their runtime costs. In this work, we approach this challenge from a causal inference perspective, and we propose a scoring mechanism to facilitate structured pruning of NNs. The approach is based on measuring mutual information under a maximum entropy perturbation, sequentially propagated through the NN. We demonstrate the method's performance on two datasets and various NNs' sizes, and we show that our approach achieves competitive performance under challenging conditions.

Anush Sankaran, Olivier Mastropietro, Ehsan Saboori et al.

Designing deep learning-based solutions is becoming a race for training deeper models with a greater number of layers. While a large-size deeper model could provide competitive accuracy, it creates a lot of logistical challenges and unreasonable resource requirements during development and deployment. This has been one of the key reasons for deep learning models not being excessively used in various production environments, especially in edge devices. There is an immediate requirement for optimizing and compressing these deep learning models, to enable on-device intelligence. In this research, we introduce a black-box framework, Deeplite Neutrino for production-ready optimization of deep learning models. The framework provides an easy mechanism for the end-users to provide constraints such as a tolerable drop in accuracy or target size of the optimized models, to guide the whole optimization process. The framework is easy to include in an existing production pipeline and is available as a Python Package, supporting PyTorch and Tensorflow libraries. The optimization performance of the framework is shown across multiple benchmark datasets and popular deep learning models. Further, the framework is currently used in production and the results and testimonials from several clients are summarized.

Shafigh Parsazad, Ehsan Saboori, Amin Allahyar

In the most intrusion detection systems (IDS), a system tries to learn characteristics of different type of attacks by analyzing packets that sent or received in network. These packets have a lot of features. But not all of them is required to be analyzed to detect that specific type of attack. Detection speed and computational cost is another vital matter here, because in these types of problems, datasets are very huge regularly. In this paper we tried to propose a very simple and fast feature selection method to eliminate features with no helpful information on them. Result faster learning in process of redundant feature omission. We compared our proposed method with three most successful similarity based feature selection algorithm including Correlation Coefficient, Least Square Regression Error and Maximal Information Compression Index. After that we used recommended features by each of these algorithms in two popular classifiers including: Bayes and KNN classifier to measure the quality of the recommendations. Experimental result shows that although the proposed method can't outperform evaluated algorithms with high differences in accuracy, but in computational cost it has huge superiority over them.

Ehsan Saboori, Shafigh Parsazad, Anoosheh Sadeghi

The k-means algorithm is one of the well-known and most popular clustering algorithms. K-means seeks an optimal partition of the data by minimizing the sum of squared error with an iterative optimization procedure, which belongs to the category of hill climbing algorithms. As we know hill climbing searches are famous for converging to local optimums. Since k-means can converge to a local optimum, different initial points generally lead to different convergence cancroids, which makes it important to start with a reasonable initial partition in order to achieve high quality clustering solutions. However, in theory, there exist no efficient and universal methods for determining such initial partitions. In this paper we tried to find an optimum initial partitioning for k-means algorithm. To achieve this goal we proposed a new improved version of downhill simplex search, and then we used it in order to find an optimal result for clustering approach and then compare this algorithm with Genetic Algorithm base (GA), Genetic K-Means (GKM), Improved Genetic K-Means (IGKM) and k-means algorithms.

Ehsan Saboori, Shafigh Parsazad, Yasaman Sanatkhani

Network intrusion detection systems have become a crucial issue for computer systems security infrastructures. Different methods and algorithms are developed and proposed in recent years to improve intrusion detection systems. The most important issue in current systems is that they are poor at detecting novel anomaly attacks. These kinds of attacks refer to any action that significantly deviates from the normal behaviour which is considered intrusion. This paper proposed a model to improve this problem based on data mining techniques. Apriori algorithm is used to predict novel attacks and generate real-time rules for firewall. Apriori algorithm extracts interesting correlation relationships among large set of data items. This paper illustrates how to use Apriori algorithm in intrusion detection systems to cerate a automatic firewall rules generator to detect novel anomaly attack. Apriori is the best-known algorithm to mine association rules. This is an innovative way to find association rules on large scale.

Ehsan Saboori, Shahriar Mohammadi

One of the most important issues in peer-to-peer networks is anonymity. The major anonymity for peer-to-peer users concerned with the users' identities and actions which can be revealed by any other members. There are many approaches proposed to provide anonymous peer-to-peer communications. An intruder can get information about the content of the data, the sender's and receiver's identities. Anonymous approaches are designed with the following three goals: to protect the identity of provider, to protect the identity of requester and to protect the contents of transferred data between them. This article presents a new peer-to-peer approach to achieve anonymity between a requester and a provider in peer-to-peer networks with trusted servers called suppernode so that the provider will not be able to identify the requester and no other peers can identify the two communicating parties with certainty. This article shows that the proposed algorithm improved reliability and has more security. This algorithm, based on onion routing and randomization, protects transferring data against traffic analysis attack. The ultimate goal of this anonymous communications algorithm is to allow a requester to communicate with a provider in such a manner that nobody can determine the requester's identity and the content of transferred data.

Shafigh Parsazad, Ehsan Saboori, Amin Allahyar

The real challenge in pattern recognition task and machine learning process is to train a discriminator using labeled data and use it to distinguish between future data as accurate as possible. However, most of the problems in the real world have numerous data, which labeling them is a cumbersome or even an impossible matter. Semi-supervised learning is one approach to overcome these types of problems. It uses only a small set of labeled with the company of huge remain and unlabeled data to train the discriminator. In semi-supervised learning, it is very essential that which data is labeled and depend on position of data it effectiveness changes. In this paper, we proposed an evolutionary approach called Artificial Immune System (AIS) to determine which data is better to be labeled to get the high quality data. The experimental results represent the effectiveness of this algorithm in finding these data points.