arXiv subjects

Charles Fleming

Publications and source records attributed to Charles Fleming.

48 records · Page 3Linked to original sources

arXiv2025

Enhancing Dance-to-Music Generation via Negative Conditioning Latent Diffusion Model

Conditional diffusion models have gained increasing attention since their impressive results for cross-modal synthesis, where the strong alignment between conditioning input and generated output can be achieved by training a time-conditioned U-Net augmented with cross-attention mechanism. In this paper, we focus on the problem of generating music synchronized with rhythmic visual cues of the given dance video. Considering that bi-directional guidance is more beneficial for training a diffusion model, we propose to enhance the quality of generated music and its synchronization with dance videos by adopting both positive rhythmic information and negative ones (PN-Diffusion) as conditions, where a dual diffusion and reverse processes is devised. Specifically, to train a sequential multi-modal U-Net structure, PN-Diffusion consists of a noise prediction objective for positive conditioning and an additional noise prediction objective for negative conditioning. To accurately define and select both positive and negative conditioning, we ingeniously utilize temporal correlations in dance videos, capturing positive and negative rhythmic cues by playing them forward and backward, respectively. Through subjective and objective evaluations of input-output correspondence in terms of dance-music beat alignment and the quality of generated music, experimental results on the AIST++ and TikTok dance video datasets demonstrate that our model outperforms SOTA dance-to-music generation models.

cs.SD↗

arXiv2025

ProDiF: Protecting Domain-Invariant Features to Secure Pre-Trained Models Against Extraction

Pre-trained models are valuable intellectual property, capturing both domain-specific and domain-invariant features within their weight spaces. However, model extraction attacks threaten these assets by enabling unauthorized source-domain inference and facilitating cross-domain transfer via the exploitation of domain-invariant features. In this work, we introduce **ProDiF**, a novel framework that leverages targeted weight space manipulation to secure pre-trained models against extraction attacks. **ProDiF** quantifies the transferability of filters and perturbs the weights of critical filters in unsecured memory, while preserving actual critical weights in a Trusted Execution Environment (TEE) for authorized users. A bi-level optimization further ensures resilience against adaptive fine-tuning attacks. Experimental results show that **ProDiF** reduces source-domain accuracy to near-random levels and decreases cross-domain transferability by 74.65\%, providing robust protection for pre-trained models. This work offers comprehensive protection for pre-trained DNN models and highlights the potential of weight space manipulation as a novel approach to model security.

cs.CR↗

arXiv2024

Self-Adapting Large Visual-Language Models to Edge Devices across Visual Modalities

Recent advancements in Vision-Language (VL) models have sparked interest in their deployment on edge devices, yet challenges in handling diverse visual modalities, manual annotation, and computational constraints remain. We introduce EdgeVL, a novel framework that bridges this gap by seamlessly integrating dual-modality knowledge distillation and quantization-aware contrastive learning. This approach enables the adaptation of large VL models, like CLIP, for efficient use with both RGB and non-RGB images on resource-limited devices without the need for manual annotations. EdgeVL not only transfers visual language alignment capabilities to compact models but also maintains feature quality post-quantization, significantly enhancing open-vocabulary classification performance across various visual modalities. Our work represents the first systematic effort to adapt large VL models for edge deployment, showcasing up to 15.4% accuracy improvements on multiple datasets and up to 93-fold reduction in model size.

cs.CV↗

arXiv2024

Quantum Secure Anonymous Communication Networks

Anonymous communication networks (ACNs) enable Internet browsing in a way that prevents the accessed content from being traced back to the user. This allows a high level of privacy, protecting individuals from being tracked by advertisers or governments, for example. The Tor network, a prominent example of such a network, uses a layered encryption scheme to encapsulate data packets, using Tor nodes to obscure the routing process before the packets enter the public Internet. While Tor is capable of providing substantial privacy, its encryption relies on schemes, such as RSA and Diffie-Hellman for distributing symmetric keys, which are vulnerable to quantum computing attacks and are currently in the process of being phased out. To overcome the threat, we propose a quantum-resistant alternative to RSA and Diffie-Hellman for distributing symmetric keys, namely, quantum key distribution (QKD). Standard QKD networks depend on trusted nodes to relay keys across long distances, however, reliance on trusted nodes in the quantum network does not meet the criteria necessary for establishing a Tor circuit in the ACN. We address this issue by developing a protocol and network architecture that integrates QKD without the need for trusted nodes, thus meeting the requirements of the Tor network and creating a quantum-secure anonymous communication network.

quant-ph↗

arXiv2023

Privacy Threats in Stable Diffusion Models

This paper introduces a novel approach to membership inference attacks (MIA) targeting stable diffusion computer vision models, specifically focusing on the highly sophisticated Stable Diffusion V2 by StabilityAI. MIAs aim to extract sensitive information about a model's training data, posing significant privacy concerns. Despite its advancements in image synthesis, our research reveals privacy vulnerabilities in the stable diffusion models' outputs. Exploiting this information, we devise a black-box MIA that only needs to query the victim model repeatedly. Our methodology involves observing the output of a stable diffusion model at different generative epochs and training a classification model to distinguish when a series of intermediates originated from a training sample or not. We propose numerous ways to measure the membership features and discuss what works best. The attack's efficacy is assessed using the ROC AUC method, demonstrating a 60\% success rate in inferring membership information. This paper contributes to the growing body of research on privacy and security in machine learning, highlighting the need for robust defenses against MIAs. Our findings prompt a reevaluation of the privacy implications of stable diffusion models, urging practitioners and developers to implement enhanced security measures to safeguard against such attacks.

cs.CV↗

arXiv2022

Scale Invariant Privacy Preserving Video via Wavelet Decomposition

Video surveillance has become ubiquitous in the modern world. Mobile devices, surveillance cameras, and IoT devices, all can record video that can violate our privacy. One proposed solution for this is privacy-preserving video, which removes identifying information from the video as it is produced. Several algorithms for this have been proposed, but all of them suffer from scale issues: in order to sufficiently anonymize near-camera objects, distant objects become unidentifiable. In this paper, we propose a scale-invariant method, based on wavelet decomposition.

cs.CR↗

arXiv2022

Focused Adversarial Attacks

Recent advances in machine learning show that neural models are vulnerable to minimally perturbed inputs, or adversarial examples. Adversarial algorithms are optimization problems that minimize the accuracy of ML models by perturbing inputs, often using a model's loss function to craft such perturbations. State-of-the-art object detection models are characterized by very large output manifolds due to the number of possible locations and sizes of objects in an image. This leads to their outputs being sparse and optimization problems that use them incur a lot of unnecessary computation. We propose to use a very limited subset of a model's learned manifold to compute adversarial examples. Our \textit{Focused Adversarial Attacks} (FA) algorithm identifies a small subset of sensitive regions to perform gradient-based adversarial attacks. FA is significantly faster than other gradient-based attacks when a model's manifold is sparsely activated. Also, its perturbations are more efficient than other methods under the same perturbation constraints. We evaluate FA on the COCO 2017 and Pascal VOC 2007 detection datasets.

cs.LG↗

arXiv2022

Ulixes: Facial Recognition Privacy with Adversarial Machine Learning

Facial recognition tools are becoming exceptionally accurate in identifying people from images. However, this comes at the cost of privacy for users of online services with photo management (e.g. social media platforms). Particularly troubling is the ability to leverage unsupervised learning to recognize faces even when the user has not labeled their images. In this paper we propose Ulixes, a strategy to generate visually non-invasive facial noise masks that yield adversarial examples, preventing the formation of identifiable user clusters in the embedding space of facial encoders. This is applicable even when a user is unmasked and labeled images are available online. We demonstrate the effectiveness of Ulixes by showing that various classification and clustering methods cannot reliably label the adversarial examples we generate. We also study the effects of Ulixes in various black-box settings and compare it to the current state of the art in adversarial machine learning. Finally, we challenge the effectiveness of Ulixes against adversarially trained models and show that it is robust to countermeasures.

cs.CV↗

arXiv2020

The Smart$^2$ Speaker Blocker: An Open-Source Privacy Filter for Connected Home Speakers

The popularity and projected growth of in-home smart speaker assistants, such as Amazon's Echo, has raised privacy concerns among consumers and privacy advocates. Notable questions regarding the collection and storage of user data by for-profit organizations include: what data is being collected and how is it being used, who has or can obtain access to such data, and how can user privacy be maintained while providing useful services. In addition to concerns regarding what the speaker manufacturer will do with your data, there are also more fundamental concerns about the security of these devices, third-party plugins, and the servers where they store recorded data. To address these privacy and security concerns, we introduce an intermediary device to provide an additional layer of security, which we call the \textit{smart, smart speaker blocker} or Smart\textsuperscript{2} for short. By intelligently filtering sensitive conversations, and completely blocking this information from reaching a smart speaker's microphone(s), the Smart$^2$ Speaker Blocker is an open-source, network-local (offline) smart device that provides users with decisive control over what data leaves their living room.

cs.CR↗

arXiv2019

SemanticLock: An authentication method for mobile devices using semantically-linked images

We introduce SemanticLock, a single factor graphical authentication solution for mobile devices. SemanticLock uses a set of graphical images as password tokens that construct a semantically memorable story representing the user`s password. A familiar and quick action of dragging or dropping the images into their respective positions either in a \textit{continous flow} or in \textit{discrete} movements on the the touchscreen is what is required to use our solution. The authentication strength of the SemanticLock is based on the large number of possible semantic constructs derived from the positioning of the image tokens and the type of images selected. Semantic Lock has a high resistance to smudge attacks and it equally exhibits a higher level of memorability due to its graphical paradigm. In a three weeks user study with 21 participants comparing SemanticLock against other authentication systems, we discovered that SemanticLock outperformed the PIN and matched the PATTERN both on speed, memorability, user acceptance and usability. Furthermore, qualitative test also show that SemanticLock was rated more superior in like-ability. SemanticLock was also evaluated while participants walked unencumbered and walked encumbered carrying "everyday" items to analyze the effects of such activities on its usage.

cs.CR↗

arXiv2019

Lessons Learned from Developing a Microservice Based Mobile Location-Based Crowdsourcing Platform

Research in Mobile Location-Based Crowdsourcing is hindered by a marked lack of real-world data. The development of a standardized, lightweight, easily deployable, modular, composable, and most of all, scalable experimentation framework would go a long way in facilitating such research. Conveniently, these are all salient characteristics of systems developed using a microservices approach. We propose QRowdsource - a MLBC experimentation framework built using a distributed services architecture. In this paper, we discuss the design and development of QRowdsource, from the decomposition of functional components to the orchestration of services within the framework. We also take a look at how the advantages and disadvantages of using a microservices approach translate to our specific use case and deliberate over a number of lessons learned while developing the experimentation framework.

cs.HC↗

arXiv2016

Privacy-Preserving Human Activity Recognition from Extreme Low Resolution

Privacy protection from surreptitious video recordings is an important societal challenge. We desire a computer vision system (e.g., a robot) that can recognize human activities and assist our daily life, yet ensure that it is not recording video that may invade our privacy. This paper presents a fundamental approach to address such contradicting objectives: human activity recognition while only using extreme low-resolution (e.g., 16x12) anonymized videos. We introduce the paradigm of inverse super resolution (ISR), the concept of learning the optimal set of image transformations to generate multiple low-resolution (LR) training videos from a single video. Our ISR learns different types of sub-pixel transformations optimized for the activity classification, allowing the classifier to best take advantage of existing high-resolution videos (e.g., YouTube videos) by creating multiple LR training videos tailored for the problem. We experimentally confirm that the paradigm of inverse super resolution is able to benefit activity recognition from extreme low-resolution videos.

cs.CV↗