A naukri.com initiative
ML News
Arxiv
1h
145
Image Credit: Arxiv
Whispering Under the Eaves: Protecting User Privacy Against Commercial and LLM-powered Automatic Speech Recognition Systems
- The paper discusses the issue of user privacy in automatic speech recognition (ASR) systems.
- It proposes a framework called AudioShield for protecting live users' speech against ASR systems.
- Central to the framework is the concept of Transferable Universal Adversarial Perturbations in the Latent Space (LS-TUAP).
- Comprehensive evaluation demonstrates the superiority of AudioShield in protecting user privacy and improving audio quality.
Read Full Article
8 Likes
Arxiv
1h
219
Image Credit: Arxiv
Investigating the Capabilities and Limitations of Machine Learning for Identifying Bias in English Language Data with Information and Heritage Professionals
- ML systems continue to harm marginalized people despite efforts to mitigate biases.
- Creating models to identify biased language instead of removing biases is a more feasible approach.
- Limitations of ML in identifying bias include its contextual nature and potential to privilege or oppress different communities.
- Expanding ML approaches to bias and fairness is necessary to address the limitations and promote a mixed-methods investigation.
Read Full Article
13 Likes
Arxiv
1h
103
Image Credit: Arxiv
CrackSQL: A Hybrid SQL Dialect Translation System Powered by Large Language Models
- CrackSQL is a hybrid SQL dialect translation system designed to enable seamless interaction across different database systems.
- It combines rule-based and large language model (LLM)-based methods to improve translation accuracy and overcome challenges such as syntactic discrepancies and semantic variations.
- CrackSQL leverages LLMs to minimize manual intervention and introduces a cross-dialect syntax embedding model for precise syntax alignment.
- It offers multiple deployment and access options, including a web console interface, a PyPI package, and a command-line prompt.
Read Full Article
6 Likes
Arxiv
1h
196
Image Credit: Arxiv
Privacy-Preserving Transfer Learning for Community Detection using Locally Distributed Multiple Networks
- This paper presents TransNet, a new method for transfer learning in community detection of network data.
- TransNet aims to improve the clustering performance of the target network by utilizing auxiliary source networks that are privacy-preserved and locally stored across various sources.
- To achieve privacy preservation, the edges of each locally stored network are perturbed using the randomized response mechanism, ensuring differential privacy.
- By proposing an adaptive weighting method and regularization technique, TransNet effectively aggregates the eigenspaces of the source networks, incorporating the effects of privacy and heterogeneity.
Read Full Article
11 Likes
Arxiv
1h
151
Image Credit: Arxiv
Explorable INR: An Implicit Neural Representation for Ensemble Simulation Enabling Efficient Spatial and Parameter Exploration
- Researchers have proposed a surrogate model called Explorable INR.
- Explorable INR is an implicit neural representation-based model.
- It enables efficient spatial and parameter exploration.
- The model reduces computation and memory costs while providing effective ensemble analysis.
Read Full Article
9 Likes
Arxiv
1h
158
Image Credit: Arxiv
Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents
- Agent S2 is a compositional generalist-specialist framework for computer use agents.
- It automates digital tasks by interacting with graphical user interfaces (GUIs) on computers and mobile devices.
- Agent S2 overcomes challenges such as imprecise grounding and long-horizon task planning.
- It establishes new state-of-the-art performance on prominent computer use benchmarks.
Read Full Article
9 Likes
Arxiv
1h
142
Image Credit: Arxiv
AI Judges in Design: Statistical Perspectives on Achieving Human Expert Equivalence With Vision-Language Models
- Recent advances in vision-language models (VLMs) offer the potential to automate design assessments, but it is crucial to ensure that these AI ``judges'' perform on par with human experts.
- A statistical framework has been introduced to determine whether an AI judge's ratings match those of human experts in design evaluation.
- The top-performing AI judge using text- and image-based in-context learning achieves expert-level agreement for uniqueness and drawing quality and outperforms or matches trained novices across all metrics.
- Reasoning-supported VLM models can achieve human-expert equivalence in design evaluation, impacting design evaluation in education and practice.
Read Full Article
8 Likes
Arxiv
1h
242
Image Credit: Arxiv
QSViT: A Methodology for Quantizing Spiking Vision Transformers
- Vision Transformer (ViT)-based models have shown state-of-the-art performance in vision-based AI tasks.
- Spiking Vision Transformer (SViT)-based models have emerged as low-power ViT networks.
- QSViT is a methodology to compress SViT models through quantization.
- QSViT achieves memory and power savings while maintaining high accuracy.
Read Full Article
14 Likes
Arxiv
1h
193
Image Credit: Arxiv
SentenceKV: Efficient LLM Inference via Sentence-Level Semantic KV Caching
- Efficient LLM inference can be achieved through SentenceKV, a novel sentence-level semantic KV caching approach.
- SentenceKV addresses the limitations of traditional token-level caching methods by considering semantic relationships between tokens.
- By compressing sentence representations into concise semantic vectors, stored on the GPU, SentenceKV reduces memory overhead and improves computational efficiency.
- Extensive evaluations show that SentenceKV outperforms existing methods in terms of efficiency, memory usage, and model accuracy.
Read Full Article
11 Likes
Arxiv
1h
290
Image Credit: Arxiv
When To Solve, When To Verify: Compute-Optimal Problem Solving and Generative Verification for LLM Reasoning
- Scaling test-time compute has emerged as a key strategy for enhancing the reasoning capabilities of large language models (LLMs), particularly in tasks like mathematical problem-solving.
- Recent advancements in Generative Reward Models (GenRM) reframe verification as a next-token prediction task, enabling inference-time scaling along a new axis.
- However, the evaluation shows that Self-Consistency (SC) is more compute-efficient than GenRM for most practical inference budgets across diverse models and datasets.
- The work provides practical guidance on optimizing test-time scaling by balancing solution generation and verification.
Read Full Article
17 Likes
Arxiv
1h
296
Image Credit: Arxiv
Data-Driven Safety Verification using Barrier Certificates and Matrix Zonotopes
- Ensuring safety in cyber-physical systems (CPSs) is a critical challenge when system models are difficult to obtain or cannot be fully trusted.
- A data-driven safety verification framework is proposed that leverages matrix zonotopes and barrier certificates to verify system safety directly from noisy data.
- Instead of relying on a single unreliable model, a set of models is constructed that captures all possible system dynamics aligning with the observed data.
- The model set is compactly represented using matrix zonotopes for efficient computation and propagation of uncertainty, resulting in rigorous safety guarantees without requiring an explicit system model.
Read Full Article
17 Likes
Arxiv
1h
200
Image Credit: Arxiv
Designing Heterogeneous GNNs with Desired Permutation Properties for Wireless Resource Allocation
- Graph neural networks (GNNs) have been designed for learning a variety of wireless policies, leveraging permutation prior.
- To satisfy complicated permutation properties in wireless policies, heterogeneous GNNs (HetGNNs) are used.
- The paper presents a systematic approach for designing HetGNNs with desired permutation properties.
- Power allocation and hybrid precoding policies are used as examples for validation through simulations.
Read Full Article
12 Likes
Arxiv
1h
219
Image Credit: Arxiv
MSCMNet: Multi-scale Semantic Correlation Mining for Visible-Infrared Person Re-Identification
- The main challenge in the Visible-Infrared Person Re-Identification (VI-ReID) task lies in how to extract discriminative features from different modalities for matching purposes.
- A Multi-scale Semantic Correlation Mining network (MSCMNet) is proposed to comprehensively exploit semantic features at multiple scales and simultaneously reduce modality information loss during feature extraction.
- The proposed MSCMNet includes three novel components: Multi-scale Information Correlation Mining Block (MIMB), quadruple-stream feature extractor (QFE), and Quadruple Center Triplet Loss (QCT).
- Extensive experiments on various datasets show that MSCMNet achieves high accuracy in Visible-Infrared Person Re-Identification.
Read Full Article
13 Likes
Arxiv
1h
161
Image Credit: Arxiv
ResNLS: An Improved Model for Stock Price Forecasting
- Stock price forecasting is a challenging task with varying degrees of dependencies between stock prices.
- The ResNLS hybrid model improves stock price prediction by emphasizing the dependencies between adjacent stock prices.
- ResNLS, composed of ResNet and LSTM, extracts features and analyzes time series data to capture dependencies.
- ResNLS-5, using closing price data for the previous 5 trading days as input, outperforms baselines with at least a 20% improvement.
Read Full Article
9 Likes
Arxiv
1h
222
Image Credit: Arxiv
SVInvNet: A Densely Connected Encoder-Decoder Architecture for Seismic Velocity Inversion
- Researchers have developed a deep learning-based approach, SVInvNet, for seismic velocity inversion.
- SVInvNet employs a novel architecture with a multi-connection encoder-decoder structure enhanced with dense blocks.
- The model effectively processes time series data and addresses non-linear seismic velocity inversion challenges.
- Despite having fewer parameters, SVInvNet outperforms the baseline model in terms of performance.
Read Full Article
13 Likes
For uninterrupted reading, download the app