A naukri.com initiative
ML News
Arxiv
6d
111
Image Credit: Arxiv
Navigating the Future of Federated Recommendation Systems with Foundation Models
- Federated Recommendation Systems (FRSs) combined with Foundation Models (FMs) have the potential to improve client-side personalization and communication efficiency.
- The integration of FRSs and FMs can address challenges such as data sparsity and heterogeneity in client environments.
- Privacy-security trade-offs, non-IID data, and resource constraints are among the challenges introduced by this integration.
- Research directions proposed in this position paper include multimodal recommendation, real-time FM adaptation, and explainable federated reasoning.
Read Full Article
6 Likes
Arxiv
6d
168
Image Credit: Arxiv
LearnedKV: Integrating LSM and Learned Index for Superior Performance on Storage
- LearnedKV is a tiered key-value store that combines a Log-Structured Merge (LSM) tree with a Learned Index for improved read and write performance on storage systems.
- The design of LearnedKV includes a two-tier approach where the LSM tree handles recent write operations, while a separate Learned Index enhances read performance.
- A non-blocking conversion mechanism efficiently transforms LSM data into a Learned Index during garbage collection, ensuring high performance without interrupting operations.
- Through its tiered approach, LearnedKV significantly reduces LSM size, leading to substantial performance gains in both read and write operations compared to existing LSM-based solutions.
Read Full Article
10 Likes
Arxiv
6d
209
Image Credit: Arxiv
Deep Inverse Design for High-Level Synthesis
- High-level synthesis (HLS) is a key automation technique for digital circuit design.
- However, the need for expertise and time in pragma tuning remains challenging for HLS.
- To address this, a novel approach called Deep Inverse Design for HLS (DID4HLS) is proposed.
- DID4HLS integrates graph neural networks and generative models to optimize hardware designs for compute-intensive algorithms.
Read Full Article
12 Likes
Arxiv
6d
164
Image Credit: Arxiv
Online SLA Decomposition: Enabling Real-Time Adaptation to Evolving Network Systems
- A study proposes an online learning-decomposition framework to dynamically decompose Service Level Agreements (SLAs) in network slice management.
- The framework continuously updates risk models based on the most recent feedback using components like online gradient descent and FIFO memory buffers.
- Empirical study shows that the proposed framework outperforms static approaches, providing more accurate and resilient SLA decomposition.
- Comprehensive complexity analysis of the solution is also provided.
Read Full Article
9 Likes
Arxiv
6d
320
Image Credit: Arxiv
The AI Risk Repository: A Comprehensive Meta-Review, Database, and Taxonomy of Risks From Artificial Intelligence
- The AI Risk Repository is a comprehensive and categorized database, aiming to provide a common understanding of risks posed by Artificial Intelligence (AI).
- It consists of 777 risks extracted from 43 taxonomies, which can be filtered based on two overarching taxonomies.
- The repository includes a high-level causal taxonomy classifying risks based on factors like entity, intentionality, and timing, along with a mid-level domain taxonomy classifying risks into seven AI risk domains and 23 subdomains.
- By curating, analyzing, and extracting AI risk frameworks, this repository facilitates a more coordinated and complete approach to defining, auditing, and managing the risks associated with AI systems.
Read Full Article
19 Likes
Arxiv
6d
292
Image Credit: Arxiv
Decoding Android Malware with a Fraction of Features: An Attention-Enhanced MLP-SVM Approach
- A novel framework is introduced to enhance Android malware detection and classification using an attention-enhanced Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM).
- The framework achieves an impressive accuracy of over 99% by analyzing only 47 features out of over 9,760 available in the dataset.
- The MLP component, enhanced with an attention mechanism, focuses on discriminative features and reduces the feature set to 14 components using Linear Discriminant Analysis (LDA).
- The SVM component, utilizing an RBF kernel, accurately maps the reduced components to a high-dimensional space for precise classification of malware into their respective families.
Read Full Article
17 Likes
Arxiv
6d
127
Image Credit: Arxiv
Range, not Independence, Drives Modularity in Biologically Inspired Representations
- A new research work presents a theory on when biologically inspired networks modularize their representation of source variables.
- The theory provides necessary and sufficient conditions for modularization based on the spread of support of the sources.
- The research validates the theory by applying it to various empirical studies on nonlinear neural networks.
- The results suggest alternate origins of mixed-selectivity, contributing to a better understanding of modular representations.
Read Full Article
7 Likes
Arxiv
6d
308
Image Credit: Arxiv
Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces
- A single Transformer model called Dualformer is presented, which integrates both fast and slow reasoning modes.
- Dualformer is trained on data with randomized reasoning traces, dropping different parts of the traces during training.
- At inference time, Dualformer can be configured to output only solutions (fast mode), reasoning chain and solution (slow mode), or automatically decide which mode to engage (auto mode).
- In terms of performance and computational efficiency, Dualformer outperforms corresponding baseline models, showing improved performance in maze navigation tasks and math problems.
Read Full Article
18 Likes
Arxiv
6d
296
Image Credit: Arxiv
TACO: Adversarial Camouflage Optimization on Trucks to Fool Object Detectors
- TACO is a novel framework that generates adversarial camouflage patterns on 3D vehicle models to deceive object detectors.
- It integrates differentiable rendering with a Photorealistic Rendering Network to optimize adversarial textures targeted at YOLOv8.
- Experimental evaluations demonstrate that TACO significantly degrades YOLOv8's detection performance and exhibits transferability to other object detection models.
- TACO achieves an [email protected] of 0.0099 on unseen test data.
Read Full Article
17 Likes
Arxiv
6d
341
Image Credit: Arxiv
Microfoundation Inference for Strategic Prediction
- Performative prediction is a phenomenon where the predictive model itself can influence the distribution of the target variable.
- The social impacts of predictions in machine learning are often unknown to practitioners, hindering widespread adaptation.
- A new methodology is proposed to learn the distribution map that captures the long-term impacts of predictive models on the population.
- The approach leverages optimal transport to align pre-model and post-model exposure distributions.
Read Full Article
20 Likes
Arxiv
6d
12
Image Credit: Arxiv
7B Fully Open Source Moxin-LLM -- From Pretraining to GRPO-based Reinforcement Learning Enhancement
- Moxin-LLM is a fully open-source Large Language Model (LLM) adhering to principles of open science, open source, open data, and open access.
- Moxin-LLM aims to address transparency concerns by releasing pre-training code, configurations, datasets, and checkpoints, allowing further innovations on LLMs.
- Moxin-LLM goes through several finetuning stages to enhance reasoning capability, utilizing a post-training framework, instruction data, and Group Relative Policy Optimization (GRPO).
- Experiments show that Moxin-LLM achieves superior performance in zero-shot evaluation, few-shot evaluation, and Chain-of-Thought (CoT) evaluation.
Read Full Article
Like
Arxiv
6d
407
Image Credit: Arxiv
Opinion de-polarization of social networks with GNNs
- Social media is highly polarized with echo chamber structures, where users form connections only with similar-minded individuals.
- A study proposes an algorithm to decrease polarization by identifying a set of users who adopt a moderate opinion on a topic.
- The algorithm utilizes a Graph Neural Network, making it more effective for handling large graphs.
Read Full Article
24 Likes
Arxiv
6d
345
Image Credit: Arxiv
AIArena: A Blockchain-Based Decentralized AI Training Platform
- The rapid advancement of AI has highlighted challenges in its development and implementation, primarily due to centralized control by major corporations.
- AIArena is a blockchain-based decentralized AI training platform aimed at democratizing AI development and alignment through on-chain incentive mechanisms.
- AIArena fosters an open and collaborative environment, allowing participants to contribute models and computing resources.
- The evaluation of AIArena on the public Base blockchain Sepolia testnet demonstrates its feasibility in real-world applications.
Read Full Article
20 Likes
Arxiv
6d
172
Image Credit: Arxiv
Functional connectomes of neural networks
- The study of the functional connectome, which maps the functional connections between different brain regions, has provided valuable insights.
- Neural networks, inspired by the brain's architecture, have achieved notable success in diverse applications but are often noted for their lack of interpretability.
- A novel approach is proposed to bridge neural networks and human brain functions through brain-inspired techniques.
- The approach enhances the interpretability of neural networks, providing a deeper understanding of their underlying mechanisms.
Read Full Article
10 Likes
Arxiv
6d
287
Image Credit: Arxiv
Efficient and Responsible Adaptation of Large Language Models for Robust and Equitable Top-k Recommendations
- Efficient and Responsible Adaptation of Large Language Models for Robust and Equitable Top-k Recommendations
- Conventional recommendation systems (RSs) often overlook the needs of diverse user populations, leading to performance disparities and reduced robustness to sub-populations.
- A hybrid task allocation framework is proposed to promote social good and serve all user groups equitably by efficiently adapting large language models (LLMs).
- The framework involves identifying weak and inactive users, using an in-context learning approach, and evaluating the performance on real-world datasets with positive results.
Read Full Article
17 Likes
For uninterrupted reading, download the app