A naukri.com initiative
ML News
Arxiv
3d
289
Image Credit: Arxiv
On Model Protection in Federated Learning against Eavesdropping Attacks
- This study investigates the protection offered by federated learning algorithms against eavesdropping adversaries.
- The focus of the research is on safeguarding the client model itself.
- The study examines various factors that impact the level of protection, such as client selection, local objective functions, global aggregation, and eavesdropper's capabilities.
- The results highlight the limitations of methods based on differential privacy in this context.
Read Full Article
17 Likes
Arxiv
3d
136
Image Credit: Arxiv
Robust Channel Estimation for Optical Wireless Communications Using Neural Network
- This paper presents a robust channel estimation framework for optical wireless communications using a neural network.
- The framework addresses frequency selective channels to improve system reliability and performance.
- The neural network estimates general optical wireless channels without prior channel information.
- Simulation results show improved and robust performance compared to conventional methods, highlighting the potential of neural networks in enhancing OWC systems.
Read Full Article
8 Likes
Arxiv
3d
277
Image Credit: Arxiv
OmniCellTOSG: The First Cell Text-Omic Signaling Graphs Dataset for Joint LLM and GNN Modeling
- OmniCellTOSG is the first dataset of cell text-omic signaling graphs (TOSGs), which represents the signaling network of individual or meta-cells and is labeled with information such as organ, disease, sex, age, and cell subtype.
- The dataset integrates human-readable annotations and quantitative gene and protein abundance data, enabling graph reasoning to decode cell signaling.
- It is built from single-cell RNA sequencing data of approximately 120 million cells from diverse tissues and conditions, and is compatible with PyTorch.
- The OmniCellTOSG dataset has the potential to transform research in life sciences, healthcare, and precision medicine.
Read Full Article
16 Likes
Arxiv
3d
92
Image Credit: Arxiv
Less-to-More Generalization: Unlocking More Controllability by In-Context Generation
- Subject-driven generation in image generation faces challenges in data scalability and subject expansibility.
- A data synthesis pipeline, utilizing in-context generation capabilities, is proposed to address these challenges.
- UNO, a multi-image conditioned subject-to-image model, is introduced for controllable and consistent generation.
- Experiments demonstrate the effectiveness of the proposed method in single-subject and multi-subject driven generation.
Read Full Article
5 Likes
Arxiv
3d
124
Image Credit: Arxiv
MDP: Multidimensional Vision Model Pruning with Latency Constraint
- Current structural pruning methods face limitations in aggressive parameter reduction and latency-aware optimization.
- Multi-Dimensional Pruning (MDP) addresses these limitations by optimizing across various granularities and using advanced latency modeling.
- MDP achieves an optimal balance between latency and accuracy by formulating pruning as a Mixed-Integer Nonlinear Program (MINLP).
- Experimental results show that MDP outperforms previous methods, achieving speed increase and accuracy improvement.
Read Full Article
7 Likes
Arxiv
3d
217
Image Credit: Arxiv
Example-Free Learning of Regular Languages with Prefix Queries
- Language learning refers to the problem of inferring a mathematical model which represents a formal language.
- Existing language learning algorithms do not make use of prefix queries, which provide additional information about where parsing failed.
- PL* is a novel language learning algorithm that uses prefix queries, improving efficiency and accuracy compared to the classical L* algorithm.
- PL* can accurately learn a range of languages of practical interest in a more constrained setting with only prefix queries available.
Read Full Article
13 Likes
Arxiv
3d
233
Image Credit: Arxiv
FastFlow: Early Yet Robust Network Flow Classification using the Minimal Number of Time-Series Packets
- FastFlow is a time-series flow classification method that accurately classifies network flows as known types or unknown types.
- FastFlow dynamically selects the minimal number of packets to balance accuracy and efficiency.
- The method utilizes a flow representation process and a sequential decision-based classification model trained with reinforcement learning.
- FastFlow achieved superior performance in early and accurate flow classification in evaluations on public datasets.
Read Full Article
14 Likes
Arxiv
3d
120
Image Credit: Arxiv
FT-Transformer: Resilient and Reliable Transformer with End-to-End Fault Tolerant Attention
- Researchers propose an error-resilient framework called end-to-end fault tolerant attention (EFTA) for Transformer models.
- EFTA incorporates error detection and correction within a fully fused attention kernel, reducing redundant data access and mitigating memory faults.
- The framework introduces architecture-aware algorithm-based fault tolerance (ABFT) using tensor checksum to minimize communication overhead during error detection.
- Experimental results show that EFTA achieves up to 7.56x speedup over traditional methods with an average fault tolerance overhead of 13.9%.
Read Full Article
7 Likes
Arxiv
3d
92
Image Credit: Arxiv
Quantum Deep Sets and Sequences
- This paper introduces the quantum deep sets model, expanding the quantum machine learning tool-box by enabling the possibility of learning variadic functions using quantum systems.
- One variant focuses on mapping sets to quantum systems through state vector averaging, allowing the definition of a permutation-invariant variadic model.
- Another variant is useful for ordered sets, such as sequences, and relies on optimal coherification of tristochastic tensors that implement products of mixed states.
- The efficacy and versatility of quantum deep sets and sequences (QDSs) is demonstrated through synthetic problem examples.
Read Full Article
5 Likes
Arxiv
3d
349
Image Credit: Arxiv
MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism
- MegaScale-Infer is an efficient system for serving large-scale Mixture-of-Experts (MoE) models.
- It disaggregates attention and feed-forward network (FFN) modules within each model layer.
- MegaScale-Infer introduces ping-pong pipeline parallelism to exploit MoE's sparsity.
- Experimental results show that MegaScale-Infer achieves higher per-GPU throughput than other solutions.
Read Full Article
21 Likes
Arxiv
3d
373
Image Credit: Arxiv
Engineering Artificial Intelligence: Framework, Challenges, and Future Direction
- Over the past ten years, the application of artificial intelligence (AI) and machine learning (ML) in engineering domains has gained significant popularity.
- This paper introduces the 'ABCDE' as the key elements of Engineering AI and proposes a unified, systematic engineering AI ecosystem framework.
- The framework includes eight essential layers, along with attributes, goals, and applications, to guide the development and deployment of AI solutions for specific engineering needs.
- The paper also examines key challenges and highlights nine future research directions to advance the implementation of AI in engineering.
Read Full Article
22 Likes
Arxiv
3d
88
Image Credit: Arxiv
Beyond Conventional Transformers: The Medical X-ray Attention (MXA) Block for Improved Multi-Label Diagnosis Using Knowledge Distillation
- Medical imaging, particularly X-ray analysis, often involves detecting multiple conditions simultaneously within a single scan.
- A novel attention mechanism, the Medical X-ray Attention (MXA) block, has been introduced to improve X-ray abnormality detection.
- The MXA block integrates local information and broader global context using a specialized module.
- Employing knowledge distillation, the proposed model achieves a substantial improvement over the baseline model, achieving an AUC of 0.85.
Read Full Article
5 Likes
Arxiv
3d
217
Image Credit: Arxiv
FEASE: Shallow AutoEncoding Recommender with Cold Start Handling via Side Features
- FEASE is an augmented EASE model that addresses user and item cold starts in recommendation systems
- It seamlessly integrates user and item side information to handle cold start issues
- FEASE leverages rich content signals for cold items and refines user representations in data-sparse environments
- Experimental results show improved recommendation accuracy and robustness compared to previous collaborative filtering approaches
Read Full Article
13 Likes
Arxiv
3d
245
Image Credit: Arxiv
Causal Self-supervised Pretrained Frontend with Predictive Code for Speech Separation
- Speech separation (SS) seeks to disentangle a multi-talker speech mixture into single-talker speech streams.
- Causal separation models, which rely only on past and present information, offer a promising solution for real-time streaming.
- A novel frontend is introduced to mitigate the mismatch between training and run-time inference by incorporating future information into causal models through predictive patterns.
- The pretrained frontend employs a transformer decoder network with a causal convolutional encoder as the backbone and is pretrained in a self-supervised manner with innovative pretext tasks.
Read Full Article
14 Likes
Arxiv
3d
100
Image Credit: Arxiv
Dynamic Assortment Selection and Pricing with Censored Preference Feedback
- This study investigates dynamic multi-product selection and pricing using a censored multinomial logit (C-MNL) choice model.
- The goal is to maximize seller revenue by adjusting product offerings and prices based on buyer preferences and purchase feedback.
- The proposed approach combines a Lower Confidence Bound (LCB) pricing strategy with an Upper Confidence Bound (UCB) or Thompson Sampling (TS) product selection approach.
- Simulations validate the effectiveness of the methods in maximizing seller revenue.
Read Full Article
6 Likes
For uninterrupted reading, download the app