Mohamed Mejri

Elyes Manai, Mohamed Mejri, Jaouhar Fattahi

This paper investigates the impact of feature encoding techniques on the explainability of XAI (Explainable Artificial Intelligence) algorithms. Using a malware classification dataset, we trained an XGBoost model and compared the performance of two feature encoding methods: Label Encoding (LE) and One Hot Encoding (OHE). Our findings reveal a marginal performance loss when using OHE instead of LE. However, the more detailed explanations provided by OHE compensated for this loss. We observed that OHE enables deeper exploration of details in both global and local contexts, facilitating more comprehensive answers. Additionally, we observed that using OHE resulted in smaller explanation files and reduced analysis time for human analysts. These findings emphasize the significance of considering feature encoding techniques in XAI research and suggest potential for further exploration by incorporating additional encoding methods and innovative visualization approaches.

Mohamed Mejri, Chandramouli Amarnath, Abhijit Chatterjee

In recent years, both online and offline deep learning models have been developed for time series forecasting. However, offline deep forecasting models fail to adapt effectively to changes in time-series data, while online deep forecasting models are often expensive and have complex training procedures. In this paper, we reframe the online nonlinear time-series forecasting problem as one of linear hyperdimensional time-series forecasting. Nonlinear low-dimensional time-series data is mapped to high-dimensional (hyperdimensional) spaces for linear hyperdimensional prediction, allowing fast, efficient and lightweight online time-series forecasting. Our framework, TSF-HD, adapts to time-series distribution shifts using a novel co-training framework for its hyperdimensional mapping and its linear hyperdimensional predictor. TSF-HD is shown to outperform the state of the art, while having reduced inference latency, for both short-term and long-term time series forecasting. Our code is publicly available at http://github.com/tsfhd2024/tsf-hd.git

Mohamed Mejri, Ashiqur Rasul, Abhijit Chatterjee

Modern convolutional neural networks (CNNs) are workhorses for video and image processing, but fail to adapt to the computational complexity of input samples in a dynamic manner to minimize energy consumption. In this research, we propose DANCE, a fine-grained, input-aware, dynamic pruning framework for 3D CNNs to maximize power efficiency with negligible to zero impact on performance. In the proposed two-step approach, the first step is called activation variability amplification (AVA), and the 3D CNN model is retrained to increase the variance of the magnitude of neuron activations across the network in this step, facilitating pruning decisions across diverse CNN input scenarios. In the second step, called adaptive activation pruning (AAP), a lightweight activation controller network is trained to dynamically prune frames, channels, and features of 3D convolutional layers of the network (different for each layer), based on statistics of the outputs of the first layer of the network. Our method achieves substantial savings in multiply-accumulate (MAC) operations and memory accesses by introducing sparsity within convolutional layers. Hardware validation on the NVIDIA Jetson Nano GPU and the Qualcomm Snapdragon 8 Gen 1 platform demonstrates respective speedups of 1.37X and 2.22X, achieving up to 1.47X higher energy efficiency compared to the state of the art.

Mohamed Mejri, Chandramouli Amarnath, Abhijit Chatterjee

Human cognition excels at symbolic reasoning, deducing abstract rules from limited samples. This has been explained using symbolic and connectionist approaches, inspiring the development of a neuro-symbolic architecture that combines both paradigms. In parallel, recent studies have proposed the use of a "relational bottleneck" that separates object-level features from abstract rules, allowing learning from limited amounts of data . While powerful, it is vulnerable to the curse of compositionality meaning that object representations with similar features tend to interfere with each other. In this paper, we leverage hyperdimensional computing, which is inherently robust to such interference to build a compositional architecture. We adapt the "relational bottleneck" strategy to a high-dimensional space, incorporating explicit vector binding operations between symbols and relational representations. Additionally, we design a novel high-dimensional attention mechanism that leverages this relational representation. Our system benefits from the low overhead of operations in hyperdimensional space, making it significantly more efficient than the state of the art when evaluated on a variety of test datasets, while maintaining higher or equal accuracy.

Mohamed Mejri, Chandramouli Amarnath, Abhijit Chatterjee

Modern transformer-based encoder-decoder architectures struggle with reasoning tasks due to their inability to effectively extract relational information between input objects (data/tokens). Recent work introduced the Abstractor module, embedded between transformer layers, to address this gap. However, the Abstractor layer while excelling at capturing relational information (pure relational reasoning), faces challenges in tasks that require both object and relational-level reasoning (partial relational reasoning). To address this, we propose RESOLVE, a neuro-vector symbolic architecture that combines object-level features with relational representations in high-dimensional spaces, using fast and efficient operations such as bundling (summation) and binding (Hadamard product) allowing both object-level features and relational representations to coexist within the same structure without interfering with one another. RESOLVE is driven by a novel attention mechanism that operates in a bipolar high dimensional space, allowing fast attention score computation compared to the state-of-the-art. By leveraging this design, the model achieves both low compute latency and memory efficiency. RESOLVE also offers better generalizability while achieving higher accuracy in purely relational reasoning tasks such as sorting as well as partial relational reasoning tasks such as math problem-solving compared to state-of-the-art methods.

Mohamed Mejri, Aymen Mejri, Oumayma Mejri et al.

Breast cancer is one of the factors that cause the increase of mortality of women. The most widely used method for diagnosing this geological disease i.e. breast cancer is the ultrasound scan. Several key features such as the smoothness and the texture of the tumor captured through ultrasound scans encode the abnormality of the breast tumors (malignant from benign). However, ultrasound scans are often noisy and include irrelevant parts of the breast that may bias the segmentation of eventual tumors. In this paper, we are going to extract the region of interest ( i.e, bounding boxes of the tumors) and feed-forward them to one semantic segmentation encoder-decoder structure based on its classification (i.e, malignant or benign). the whole process aims to build an instance-based segmenter from a semantic segmenter and an object detector.

Jaouhar Fattahi, Mohamed Mejri

Fingerprint recognition is often a game-changing step in establishing evidence against criminals. However, we are increasingly finding that criminals deliberately alter their fingerprints in a variety of ways to make it difficult for technicians and automatic sensors to recognize their fingerprints, making it tedious for investigators to establish strong evidence against them in a forensic procedure. In this sense, deep learning comes out as a prime candidate to assist in the recognition of damaged fingerprints. In particular, convolution algorithms. In this paper, we focus on the recognition of damaged fingerprints by Convolutional Long Short-Term Memory networks. We present the architecture of our model and demonstrate its performance which exceeds 95% accuracy, 99% precision, and approaches 95% recall and 99% AUC.

Mohamed Mejri, Aymen Mejri

We present a novel and practical deep learning pipeline termed RandomForestMLP. This core trainable classification engine consists of a convolutional neural network backbone followed by an ensemble-based multi-layer perceptrons core for the classification task. It is designed in the context of self and semi-supervised learning tasks to avoid overfitting while training on very small datasets. The paper details the architecture of the RandomForestMLP and present different strategies for neural network decision aggregation. Then, it assesses its robustness to overfitting when trained on realistic image datasets and compares its classification performance with existing regular classifiers.

Jaouhar Fattahi, Mohamed Mejri

In this paper, we put forward a new tool, called SpaML, for spam detection using a set of supervised and unsupervised classifiers, and two techniques imbued with Natural Language Processing (NLP), namely Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF). We first present the NLP techniques used. Then, we present our classifiers and their performance on each of these techniques. Then, we present our overall Ensemble Learning classifier and the strategy we are using to combine them. Finally, we present the interesting results shown by SpaML in terms of accuracy and precision.

Mohamed Mejri, Antoine Richard, Cédric Pradalier

The analysis of the internal structure of trees is highly important for both forest experts, biological scientists, and the wood industry. Traditionally, CT-scanners are considered as the most efficient way to get an accurate inner representation of the tree. However, this method requires an important investment and reduces the cost-effectiveness of this operation. Our goal is to design neural-network-based methods to predict the internal density of the tree from its external bark shape. This paper compares different image-to-image(2D), volume-to-volume(3D) and Convolutional Long Short Term Memory based neural network architectures in the context of the prediction of the defect distribution inside trees from their external bark shape. Those models are trained on a synthetic dataset of 1800 CT-scanned look-like volumetric structures of the internal density of the trees and their corresponding external surface.

Jaouhar Fattahi, Mohamed Mejri, Emil Pricop

Witness functions have recently been introduced in cryptographic protocols' literature as a new powerful way to prove protocol correctness with respect to secrecy. In this paper, we extend them to the property of authentication. We show how to use them safely and we run an analysis on a modified version of the Woo-Lam protocol. We show that it is correct with respect to authentication.

Jaouhar Fattahi, Mohamed Mejri, Emil Pricop

In this paper, we show how practical the little theorem of witness functions is in detecting security flaws in some category of cryptographic protocols. We convey a formal analysis of the Needham-Schroeder symmetric-key protocol in the theory of witness functions. We show how it helps to teach about a security vulnerability in a given step of this protocol where the value of security of a particular sensitive ticket in a sent message unexpectedly plummets compared with its value when received. This vulnerability may be exploited by an intruder to mount a replay attack as described by Denning and Sacco.

Jaouhar Fattahi, Mohamed Mejri, Hanane Houmani

In this paper, we look at the property of secrecy through the growth of the protocol. Intuitively, an increasing protocol preserves the secret. For that, we need functions to estimate the security of messages. Here, we give relaxed conditions on the functions and on the protocol and we prove that an increasing protocol is correct when analyzed with functions that meet these conditions.

Jaouhar Fattahi, Mohamed Mejri, Hanane Houmani

In this paper, we present a new formal method to analyze cryptographic protocols statically for the property of secrecy. It consists in inspecting the level of security of every component in the protocol and making sure that it does not diminish during its life cycle. If yes, it concludes that the protocol keeps its secret inputs. We analyze in this paper an amended version of the Woo-Lam protocol using this new method.

Jaouhar Fattahi, Mohamed Mejri

In this paper, we use the witness-functions to analyze cryptographic protocols for secrecy under nonempty equational theories. The witness-functions are safe metrics used to compute security. An analysis with a witness-function consists in making sure that the security of every atomic message does not decrease during its lifecycle in the protocol. The analysis gets more difficult under nonempty equational theories. Indeed, the intruder can take advantage of the algebraic properties of the cryptographic primitives to derive secrets. These properties arise from the use of mathematical functions, such as multiplication, addition, exclusive-or or modular exponentiation in the cryptosystems and the protocols. Here, we show how to use the witness-functions under nonempty equational theories and we run an analysis on the Needham-Schroeder-Lowe protocol under the cipher homomorphism. This analysis reveals that although this protocol is proved secure under the perfect encryption assumption, its security collapses under the homomorphic primitives. We show how the witness-functions help to illustrate an attack scenario on it and we propose an amended version to fix it.

Jaouhar Fattahi, Mohamed Mejri, Marwa Ziadia et al.

Proving that a cryptographic protocol is correct for secrecy is a hard task. One of the strongest strategies to reach this goal is to show that it is increasing, which means that the security level of every single atomic message exchanged in the protocol, safely evaluated, never deceases. Recently, two families of functions have been proposed to measure the security level of atomic messages. The first one is the family of interpretation-functions. The second is the family of witness-functions. In this paper, we show that the witness-functions are more efficient than interpretation-functions. We give a detailed analysis of an ad-hoc protocol on which the witness-functions succeed in proving its correctness for secrecy while the interpretation-functions fail to do so.

Jaouhar Fattahi, Mohamed Mejri, Marwa Ziadia et al.

In several critical military missions, more than one decision level are involved. These decision levels are often independent and distributed, and sensitive pieces of information making up the military mission must be kept hidden from one level to another even if all of the decision levels cooperate to accomplish the same task. Usually, a mission is negotiated through insecure networks such as the Internet using cryptographic protocols. In such protocols, few security properties have to be ensured. However, designing a secure cryptographic protocol that ensures several properties at once is a very challenging task. In this paper, we propose a new secure protocol for multipart military missions that involve two independent and distributed decision levels having different security levels. We show that it ensures the secrecy, authentication, and non-repudiation properties. In addition, we show that it resists against man-in-the-middle attacks.

Mohamed Mejri, Hamdi Yahyaoui

We propose in this paper the Security Policy Language (SePL), which is a formal language for capturing and integrating distributed security policies. The syntax of SePL includes several operators for the integration of policies and it is endowed with a denotational semantics that is a generic semantics, i.e., which is independent of any evaluation environment. We prove the completeness of SePL with respect to sets theory. Furthermore, we provide a formalization of a subset of the eXtensible Access Control Markup Language (XACML), which is the well-known standard informal specification language of Web security policies. We provide also a semantics for XACML policy combining algorithms.

Jaouhar Fattahi, Mohamed Mejri, Hanane Houmani

In this paper, we present a new semi-decidable procedure to analyze cryptographic protocols for secrecy based on a new class of functions that we call: the Witness-Functions. A Witness-Function is a reliable function that guarantees the secrecy in any protocol proved increasing once analyzed by it. Hence, the problem of correctness becomes a problem of protocol growth. A Witness-Function operates on derivative messages in a role-based specification and introduces new derivation techniques. We give here the technical aspects of the Witness-Functions and we show how to use them in a semi-decidable procedure. Then, we analyze a variation of the Needham-Schroeder protocol and we show that a Witness-Function can also help to teach about flaws. Finally, we analyze the NSL protocol and we prove that it is correct with respect to secrecy.