menu
techminis

A naukri.com initiative

google-web-stories
Home

>

ML News

ML News

source image

Arxiv

2h

read

253

img
dot

Image Credit: Arxiv

Multi-Agent Path Finding in Continuous Spaces with Projected Diffusion Models

  • Multi-Agent Path Finding (MAPF) is a fundamental problem in robotics, requiring the computation of collision-free paths for multiple agents moving from their respective start to goal positions.
  • Coordinating multiple agents in a shared environment poses challenges in continuous spaces, where traditional optimization algorithms struggle with scalability.
  • Diffusion models have shown promise in single-agent path planning, but extending them to MAPF introduces new challenges for constraint feasibility, such as inter-agent collision avoidance.
  • To address this, a novel approach is proposed that combines constrained optimization with diffusion models for MAPF in continuous spaces, resulting in feasible multi-agent trajectories that respect collision avoidance and kinematic constraints.

Read Full Article

like

15 Likes

source image

Arxiv

2h

read

239

img
dot

Image Credit: Arxiv

Shifted Composition III: Local Error Framework for KL Divergence

  • This paper introduces a shifted composition rule to adapt coupling arguments to the Kullback-Leibler (KL) divergence.
  • The framework combines local error analysis and Girsanov's theorem to yield tight bounds and KL divergence guarantees.
  • It is applicable in cases of strongly log-concave, weakly log-concave, or log-Sobolev target distributions.
  • The results include KL guarantees for the randomized midpoint discretization of the Langevin diffusion.

Read Full Article

like

14 Likes

source image

Arxiv

2h

read

39

img
dot

Image Credit: Arxiv

Combinatorial Regularity for Relatively Perfect Discrete Morse Gradient Vector Fields of ReLU Neural Networks

  • A study introduces a schematic for translating between a given piecewise linear Morse function on a canonical polyhedral complex and a compatible discrete Morse function for ReLU neural networks.
  • The study expands computational tools for analyzing the topological properties of ReLU neural networks using discrete Morse theory.
  • An algorithm is introduced to determine if a given vertex in a canonical polyhedral complex corresponds to a piecewise linear Morse critical point.
  • New realizability results are provided for shallow ReLU neural networks with respect to sublevel set topology.

Read Full Article

like

2 Likes

source image

Arxiv

2h

read

302

img
dot

Image Credit: Arxiv

An information theoretic limit to data amplification

  • Generative artificial intelligence, such as Generative Adversarial Networks (GANs), has been used to amplify data for scientific analysis, allowing for data generation in reduced computing time.
  • The process of data amplification, which violates the principle of getting information for free, can result in a gain factor greater than one while keeping the information content unchanged.
  • This study presents a mathematical bound for data amplification, dependent on the number of generated and training events, and determines conditions for ensuring this bound.
  • While the resolution of variables in amplified data is not improved, the increase in sample size can improve statistical significance.

Read Full Article

like

18 Likes

source image

Arxiv

2h

read

285

img
dot

Image Credit: Arxiv

Emoji Retrieval from Gibberish or Garbled Social Media Text: A Novel Methodology and A Case Study

  • Emojis are often lost in noisy or garbled text on social media, posing challenges for data analysis and machine learning.
  • A three-step reverse-engineering methodology is proposed to retrieve emojis from garbled text in social media posts.
  • The methodology was applied to 509,248 tweets about the Mpox outbreak, retrieving 157,748 emojis from 76,914 tweets.
  • Improvements in text readability and coherence were demonstrated through various metrics, and the usage patterns of individual emojis were analyzed.

Read Full Article

like

17 Likes

source image

Arxiv

2h

read

279

img
dot

Image Credit: Arxiv

Fair Knowledge Tracing in Second Language Acquisition

  • This study evaluates the fairness of two predictive models in second-language acquisition using the Duolingo dataset.
  • Key findings include the superiority of deep learning over machine learning in terms of accuracy and fairness in second-language knowledge tracing.
  • Both models show a bias towards mobile users over non-mobile users.
  • Machine learning exhibits stronger bias against developing countries compared to deep learning.

Read Full Article

like

16 Likes

source image

Arxiv

2h

read

322

img
dot

Image Credit: Arxiv

MMFactory: A Universal Solution Search Engine for Vision-Language Tasks

  • MMFactory is a universal solution search engine for vision-language tasks.
  • It aims to provide a diverse pool of programmatic solutions by instantiating and combining visio-lingual tools from its model repository.
  • MMFactory considers user constraints and proposes solutions that meet unique design constraints.
  • Experimental results show that MMFactory outperforms existing methods in delivering tailored solutions.

Read Full Article

like

19 Likes

source image

Arxiv

2h

read

223

img
dot

Image Credit: Arxiv

Heterogeneous transfer learning for high dimensional regression with feature mismatch

  • Researchers propose a two-stage method for transferring knowledge from a source domain to a target domain in high-dimensional regression models with feature mismatch.
  • Existing transfer learning methods assume that the source and target domains have the same feature space, which limits their practical applicability.
  • The proposed method involves learning the relationship between missing and observed features in the source domain and solving a joint penalized regression optimization problem in the target domain.
  • The researchers provide an upper bound on the parameter estimation risk and prediction risk, assuming sparse differences between the source and target domain parameters.

Read Full Article

like

13 Likes

source image

Arxiv

2h

read

207

img
dot

Image Credit: Arxiv

Generating Traffic Scenarios via In-Context Learning to Learn Better Motion Planner

  • Motion planning is crucial in autonomous driving, but curating datasets for training motion planners is expensive and may not capture rare critical scenarios.
  • Researchers propose an inexpensive method for generating diverse critical traffic scenarios to train robust motion planners.
  • They use scripts to represent traffic scenarios and train a Large Language Model (LLM) to generate scripts from user-specified text descriptions.
  • Motion planners trained with the generated synthetic data outperform those trained solely on real-world data.

Read Full Article

like

12 Likes

source image

Arxiv

2h

read

118

img
dot

Image Credit: Arxiv

BRIDGE: Bundle Recommendation via Instruction-Driven Generation

  • Bundle recommendation aims to suggest a set of interconnected items to users.
  • BRIDGE is a novel framework for bundle recommendation that addresses challenges posed by diverse interaction types and sparse interaction matrices.
  • It consists of two main components: correlation-based item clustering and pseudo bundle generation.
  • Experimental results validate the superiority of BRIDGE over state-of-the-art ranking-based methods across five benchmark datasets.

Read Full Article

like

7 Likes

source image

Arxiv

2h

read

52

img
dot

Image Credit: Arxiv

Age Optimal Sampling for Unreliable Channels under Unknown Channel Statistics

  • This paper discusses the optimization of sampling policies for unreliable channels with unknown channel statistics.
  • The system consists of a sensor sending updates to a receiver through an error-prone channel, with transmission results sent back via a reliable channel.
  • The objective is to minimize the Age of Information (AoI) metric by designing a sampling policy.
  • The proposed algorithm is shown to be effective in minimizing AoI and improving stability.

Read Full Article

like

3 Likes

source image

Arxiv

2h

read

105

img
dot

Image Credit: Arxiv

An Instrumental Value for Data Production and its Application to Data Pricing

  • This paper presents an approach for capturing the instrumental value of data production processes.
  • The approach considers the context of the agent's decision-making problem and prior data or information possessed.
  • In the domain of Bayesian linear regression, the value corresponds to information gain.
  • The study shows that achieving first-best revenue for data sales depends on the seller's ability to customize data requests.

Read Full Article

like

6 Likes

source image

Arxiv

2h

read

233

img
dot

Image Credit: Arxiv

Robustness-aware Automatic Prompt Optimization

  • BATprompt (By Adversarial Training prompt) is a novel method for prompt generation designed to withstand input perturbations.
  • It uses adversarial training techniques to generate prompts that have strong performance on perturbed tasks.
  • BATprompt avoids reliance on real gradients or model parameters and leverages the advanced capabilities of Large Language Models (LLMs).
  • Experiments show that BATprompt outperforms existing prompt generation methods, delivering robustness and performance under diverse perturbation scenarios.

Read Full Article

like

14 Likes

source image

Arxiv

2h

read

285

img
dot

Image Credit: Arxiv

Quantum framework for Reinforcement Learning: integrating Markov Decision Process, quantum arithmetic, and trajectory search

  • This paper introduces a quantum framework for addressing reinforcement learning (RL) tasks, grounded in the quantum principles and leveraging a fully quantum model of the classical Markov Decision Process (MDP).
  • The implementation and optimization of agent-environment interactions are done entirely within the quantum domain, eliminating reliance on classical computations.
  • Key contributions include quantum-based state transitions, return calculation, and trajectory search mechanisms that utilize quantum principles to demonstrate the realization of RL processes through quantum phenomena.
  • Experimental results show the capacity of a quantum model to achieve quantum advantage in RL, highlighting the potential of fully quantum implementations in decision-making tasks.

Read Full Article

like

17 Likes

source image

Arxiv

2h

read

187

img
dot

Image Credit: Arxiv

U-Mamba-Net: A highly efficient Mamba-based U-net style network for noisy and reverberant speech separation

  • U-Mamba-Net is a lightweight Mamba-based U-style model for speech separation in complex environments.
  • The model aims to address the increasing size and computational load of existing speech separation models.
  • Mamba is a state space sequence model that incorporates feature selection capabilities.
  • U-Mamba-Net achieves improved performance with low computational cost according to the results on Libri2mix.

Read Full Article

like

11 Likes

For uninterrupted reading, download the app