Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 4
Encontrada(s) 1 página(s)

Dipòsit Digital de Documents de la UAB
Publicación Capítulo o parte de un libro (BookPart)

Decision tree based inference of lightning network client implementations

  • Espinasa Vilarrasa, Pol|||0009-0000-3933-4970
  • Sanvicente, Sílvia
  • Pérez-Solà, Cristina|||0000-0001-7534-1326
  • Herrera-Joancomartí, Jordi|||0000-0002-4935-4681
The Lightning Network (LN) is a second layer payment protocol on top of Bitcoin. It creates a peer-to-peer (P2P) network of payment channels that enable instant payments. The LN can be accessed through different implementations or clients, the most popular being Lightning Network Daemon (LND), Core Lightning Network (CLN), and Eclair. The first step in many known attacks to the LN is to infer the software client the node is running. This paper presents two classification models based on decision trees to infer the implementation of LN clients from either the traffic of the gossip protocol or the announced BOLT #9 features, offering a cost-free means of identification. The accuracy presented by both models in our experiments is high, ranging from 87% to 100% depending on the model and the environment where it is deployed. The application of our inference models on the LN shows a prevalence of LND clients.
Dipòsit Digital de Documents de la UAB
Publicación Capítulo o parte de un libro (BookPart)

Building Resilient AI, A Solution to Data and Model Poisoning Prevention

  • Keshavarzkalhori, Ghazaleh|||0009-0002-7758-0018
  • Pérez-Solà, Cristina|||0000-0001-7534-1326
  • Navarro-Arribas, Guillermo|||0000-0003-3535-942X
  • Herrera-Joancomartí, Jordi|||0000-0002-4935-4681
In many machine learning scenarios, training occurs outside the control of the model sponsor or the entity using the model. A growing concern in such settings revolves around model poisoning and data poisoning-how training is conducted and which data contributes to the process. This paper introduces a protective scheme against model and data poisoning attacks. Leveraging cryptographic primitives such as hashes, signature schemes, and zero-knowledge proofs, the scheme ensures the integrity of the training process. Hashing maintains the continuity of data from authenticated sensors, while signatures validate the data. In the end, zero-knowledge proofs verify the correct model computation by the entity carrying out the training process. By adopting this approach, model sponsors can securely delegate training tasks, guaranteeing the authenticity of the results. Implementation and testing demonstrate the scheme's feasibility, effectively countering data and model poisoning threats.
Dipòsit Digital de Documents de la UAB
Publicación Capítulo o parte de un libro (BookPart)

SGST, bit-depth recovery-based lossless compression for hyperspectral images

  • Xi, Zuoxin
  • Mijares i Verdú, Sebastià|||0000-0003-1038-6413
  • Serra-Sagristà, Joan|||0000-0003-4729-9292
Deep learning applications to hyperspectral image (HSI) compression have attracted significant interest in recent years. Existing deep learning methods often rely on convolutional architectures, limiting their ability to capture long-range spectral dependencies. Their use of latent feature maps also requires full decoding before any downstream analysis. To address these limitations, we propose the Spectral Group Swin Transformer (SGST), a bit-depth recovery model that explicitly separates the most and least significant bits (MSB and LSB) and employs a transformer-based structure to enhance contextual modeling. Moreover, the MSB can be directly used for preliminary downstream analysis. Experiments on the HySpecNet-11k dataset show that SGST improves reconstruction quality by over 10 dB in PSNR compared to 3D-CAE and SSCNet, with lower entropy in lossy settings. While SGST's entropy rate in lossless compression is higher than the recent LineRWKV, its significantly lower computational complexity makes it better suited for onboard applications. Code is available at: https://github.com/ZuoxinXi/SGST.
Dipòsit Digital de Documents de la UAB
Publicación Capítulo o parte de un libro (BookPart)

T2F2L, Towards Trustworthy Fairness in Federated Learning Through A Verifiable Approach

  • Keshavarzkalhori, Ghazaleh|||0009-0002-7758-0018
  • Pérez-Solà, Cristina|||0000-0001-7534-1326
  • Navarro-Arribas, Guillermo|||0000-0003-3535-942X
  • Herrera-Joancomartí, Jordi|||0000-0002-4935-4681
Federated learning (FL) has emerged as a privacypreserving alternative to traditional machine learning, enabling collaborative model training without centralizing users' data. While this decentralized approach enhances privacy, it ntroduces new trust assumptions. Specifically, that users honestly participate in the protocol. Simultaneously, the growing emphasis on fairness in machine learning has led to the development of fairness-aware FL schemes, which aim to ensure equitable model performance across diverse user groups. However, existing solutions for fair FL schemes rely on users truthfully reporting fairness-related metrics. This trust assumption opens the door to malicious behavior: users may manipulate these statistics to influence the global model, degrading its fairness, and/or its overall accuracy. Current frameworks lack mechanisms to detect or prevent such adversarial manipulation. This paper addresses this critical gap through two main contributions. First, as an example to illustrate our point, we empirically demonstrate the vulnerability of a representative fairness-aware FL framework to targeted attacks that exploit unverified fairness computations. Second, and most mportantly, we propose a novel scheme (that can be integrated into existing fair FL schemes) that augments fairness-aware FL with verifiability. Our solution enables the detection of dishonest participants without compromising user privacy, thus strengthening the robustness of fairness-aware federated learning schemes.