-
REMARK-LLM: A Robust and Efficient Watermarking Framework for Generative Large Language Models
Authors:
Ruisi Zhang,
Shehzeen Samarah Hussain,
Paarth Neekhara,
Farinaz Koushanfar
Abstract:
We present REMARK-LLM, a novel efficient, and robust watermarking framework designed for texts generated by large language models (LLMs). Synthesizing human-like content using LLMs necessitates vast computational resources and extensive datasets, encapsulating critical intellectual property (IP). However, the generated content is prone to malicious exploitation, including spamming and plagiarism.…
▽ More
We present REMARK-LLM, a novel efficient, and robust watermarking framework designed for texts generated by large language models (LLMs). Synthesizing human-like content using LLMs necessitates vast computational resources and extensive datasets, encapsulating critical intellectual property (IP). However, the generated content is prone to malicious exploitation, including spamming and plagiarism. To address the challenges, REMARK-LLM proposes three new components: (i) a learning-based message encoding module to infuse binary signatures into LLM-generated texts; (ii) a reparameterization module to transform the dense distributions from the message encoding to the sparse distribution of the watermarked textual tokens; (iii) a decoding module dedicated for signature extraction; Furthermore, we introduce an optimized beam search algorithm to guarantee the coherence and consistency of the generated content. REMARK-LLM is rigorously trained to encourage the preservation of semantic integrity in watermarked content, while ensuring effective watermark retrieval. Extensive evaluations on multiple unseen datasets highlight REMARK-LLM proficiency and transferability in inserting 2 times more signature bits into the same texts when compared to prior art, all while maintaining semantic integrity. Furthermore, REMARK-LLM exhibits better resilience against a spectrum of watermark detection and removal attacks.
△ Less
Submitted 7 April, 2024; v1 submitted 18 October, 2023;
originally announced October 2023.
-
NetFlick: Adversarial Flickering Attacks on Deep Learning Based Video Compression
Authors:
Jung-Woo Chang,
Nojan Sheybani,
Shehzeen Samarah Hussain,
Mojan Javaheripi,
Seira Hidano,
Farinaz Koushanfar
Abstract:
Video compression plays a significant role in IoT devices for the efficient transport of visual data while satisfying all underlying bandwidth constraints. Deep learning-based video compression methods are rapidly replacing traditional algorithms and providing state-of-the-art results on edge devices. However, recently developed adversarial attacks demonstrate that digitally crafted perturbations…
▽ More
Video compression plays a significant role in IoT devices for the efficient transport of visual data while satisfying all underlying bandwidth constraints. Deep learning-based video compression methods are rapidly replacing traditional algorithms and providing state-of-the-art results on edge devices. However, recently developed adversarial attacks demonstrate that digitally crafted perturbations can break the Rate-Distortion relationship of video compression. In this work, we present a real-world LED attack to target video compression frameworks. Our physically realizable attack, dubbed NetFlick, can degrade the spatio-temporal correlation between successive frames by injecting flickering temporal perturbations. In addition, we propose universal perturbations that can downgrade performance of incoming video without prior knowledge of the contents. Experimental results demonstrate that NetFlick can successfully deteriorate the performance of video compression frameworks in both digital- and physical-settings and can be further extended to attack downstream video classification networks.
△ Less
Submitted 3 April, 2023;
originally announced April 2023.
-
q-RBFNN:A Quantum Calculus-based RBF Neural Network
Authors:
Syed Saiq Hussain,
Muhammad Usman,
Taha Hasan Masood Siddique,
Imran Naseem,
Roberto Togneri,
Mohammed Bennamoun
Abstract:
In this research a novel stochastic gradient descent based learning approach for the radial basis function neural networks (RBFNN) is proposed. The proposed method is based on the q-gradient which is also known as Jackson derivative. In contrast to the conventional gradient, which finds the tangent, the q-gradient finds the secant of the function and takes larger steps towards the optimal solution…
▽ More
In this research a novel stochastic gradient descent based learning approach for the radial basis function neural networks (RBFNN) is proposed. The proposed method is based on the q-gradient which is also known as Jackson derivative. In contrast to the conventional gradient, which finds the tangent, the q-gradient finds the secant of the function and takes larger steps towards the optimal solution. The proposed $q$-RBFNN is analyzed for its convergence performance in the context of least square algorithm. In particular, a closed form expression of the Wiener solution is obtained, and stability bounds of the learning rate (step-size) is derived. The analytical results are validated through computer simulation. Additionally, we propose an adaptive technique for the time-varying $q$-parameter to improve convergence speed with no trade-offs in the steady state performance.
△ Less
Submitted 2 June, 2021;
originally announced June 2021.
-
Vehicle Intrusion And Theft Control System Using GSM and GPS -- An advance and viable approach
Authors:
Ashad Mustafa,
Hassan Jameel,
Mohtashim Baqar,
Rameez Ahmed Khan,
Zeeshan M Yaqoob,
Zeeshan Rahim,
Syed Safdar Hussain
Abstract:
This paper presents a novel approach towards the designing and development of a feasible and an embedded vehicle intrusion and theft control system using GSM (Global System for Mobile Communication) and GPS (Global Positioning System). The proposed system uses GSM technology as one of the distinguishing building blocks of the system. A GPS module Holux GR89 is used to trace the position of the veh…
▽ More
This paper presents a novel approach towards the designing and development of a feasible and an embedded vehicle intrusion and theft control system using GSM (Global System for Mobile Communication) and GPS (Global Positioning System). The proposed system uses GSM technology as one of the distinguishing building blocks of the system. A GPS module Holux GR89 is used to trace the position of the vehicle and Mercury Switches are used to collect analog data continuously, in case of an intrusion, variations will be observed in sensors reading. Continuous readings from sensors are collected on to the microcontroller constantly and on the basis of those readings decision is taken whether an intrusion is made or not and in case of an intrusion a message from a predefined set of messages is generated to the owner of the vehicle and on reception of the message, the owner will have the luxury to take an action via an SMS either to lock the gears of the vehicle or seize the engine of the vehicle from a far-off place. A relay is used to control gears and engine of the vehicle while working with the microcontroller. A prototype system was built and tested. The results were very positive and encouraging.
△ Less
Submitted 9 April, 2020;
originally announced April 2020.
-
Influence of the magnetron power on the Er-related photoluminescence of AlN:Er films prepared by magnetron sputtering
Authors:
Syed Sajjad Hussain,
Valerie Brien,
Hervé Rinnert,
Philippe Pigeat
Abstract:
The effect of magnetron power on the room temperature 1.54 $μ$m infra-red photoluminescence intensity of erbium doped AlN films grown by r. f. magnetron sputtering, has been studied. The AlN:Er thin films were deposited on (001) Silicon substrates. The study presents relative photoluminescence intensities of nanocrystallized samples prepared with identical sputtering parameters for two erbium dopi…
▽ More
The effect of magnetron power on the room temperature 1.54 $μ$m infra-red photoluminescence intensity of erbium doped AlN films grown by r. f. magnetron sputtering, has been studied. The AlN:Er thin films were deposited on (001) Silicon substrates. The study presents relative photoluminescence intensities of nanocrystallized samples prepared with identical sputtering parameters for two erbium do** levels (0.5 and 1.5 atomic %). The structural evolution of the crystallites as a function of the power is followed by transmission electron microscopy. Copyright line will be provided by the publisher 1 Introduction For some time now, rare-earth (RE)-doped semiconductors represent significant potential applications in the field of opto-electronic technology. Part of this technological interest relies on the shielded 4f levels of the RE ions as they give rise to sharp and strong luminescence peaks [1-5]. Among the RE elements, Er is preferred to its counterparts since the Er ions can produce both visible light at 558 nm (green, one of the primary colours) and IR light at 1.54 $μ$m whose spectrum region coincides with the main low-loss region in the absorption spectrum of silica-based optical fibres, combining so potential applications towards photonic devices and towards optical communication devices operating in the infrared domain. These interesting emissions can however only be exploited when placed into host matrixes. On one side, the shielding of the intra 4f levels prevents the shifting of the RE 3+ energy levels and ensures the frequency emission stability. Moreover the intra 4f transitions are parity forbidden for the isolated ions. Matrixes can render the Er 3+ ions optically active, via a relaxation of selection rules due to crystal field effects. As silicon based materials were tested in the 1960s to the 90s with no clear industrial success it was found that the
△ Less
Submitted 13 September, 2019;
originally announced September 2019.
-
Quantum Calculus-based Volterra LMS for Nonlinear Channel Estimation
Authors:
Muhammad Usman,
Muhammad Sohail Ibrahim,
Jawwad Ahmad,
Syed Saiq Hussain,
Muhammad Moinuddin
Abstract:
A novel adaptive filtering method called $q$-Volterra least mean square ($q$-VLMS) is presented in this paper. The $q$-VLMS is a nonlinear extension of conventional LMS and it is based on Jackson's derivative also known as $q$-calculus. In Volterra LMS, due to large variance of input signal the convergence speed is very low. With proper manipulation we successfully improved the convergence perform…
▽ More
A novel adaptive filtering method called $q$-Volterra least mean square ($q$-VLMS) is presented in this paper. The $q$-VLMS is a nonlinear extension of conventional LMS and it is based on Jackson's derivative also known as $q$-calculus. In Volterra LMS, due to large variance of input signal the convergence speed is very low. With proper manipulation we successfully improved the convergence performance of the Volterra LMS. The proposed algorithm is analyzed for the step-size bounds and results of analysis are verified through computer simulations for nonlinear channel estimation problem.
△ Less
Submitted 7 August, 2019;
originally announced August 2019.
-
Photoluminescence properties of Er-doped AlN films prepared by magnetron sputtering
Authors:
H. Rinnert,
S. S. Hussain,
Valerie Brien,
P. Pigeat
Abstract:
Er-doped aluminum nitride films, containing different Er concentrations, were obtained at room temperature by reactive radio frequency magnetron sputtering. The prepared samples show a nano-columnar microstructure and the size of the columns is dependent on the magnetron power. The Er-related photoluminescence (PL) was studied in relation with the temperature, the Er content and the microstructure…
▽ More
Er-doped aluminum nitride films, containing different Er concentrations, were obtained at room temperature by reactive radio frequency magnetron sputtering. The prepared samples show a nano-columnar microstructure and the size of the columns is dependent on the magnetron power. The Er-related photoluminescence (PL) was studied in relation with the temperature, the Er content and the microstructure. Steady-state PL, PL excitation spectroscopy and time-resolved PL were performed. Both visible and near infrared PL were obtained at room temperature for the as-deposited samples. It is demonstrated that the PL intensity reaches a maximum for an Er concentration equal to 1 at. % and that the PL efficiency is an increasing function of the magnetron power. Decay time measurements show the important role of defect related non radiative recombination, assumed to be correlated to the presence of grain boundaries. Moreover PL excitation results demonstrate that an indirect excitation of Er 3+ ions occurs for excitation wavelengths lower than 600 nm. It is also suggested that Er ions occupy at least two different sites in the AlN host matrix.
△ Less
Submitted 3 April, 2019;
originally announced April 2019.
