A naukri.com initiative
ML News
Arxiv
2d
360
Image Credit: Arxiv
CARIL: Confidence-Aware Regression in Imitation Learning for Autonomous Driving
- End-to-end vision-based imitation learning in autonomous driving has shown promise, but traditional approaches lack confidence estimation and precision.
- A dual-head neural network architecture is introduced that combines regression and classification heads to improve decision reliability.
- The regression head predicts continuous driving actions, while the classification head estimates confidence for corrections in low-confidence scenarios.
- Experimental results demonstrate improved driving stability, reduced lane deviation, and enhanced trajectory accuracy compared to regression-only models.
Read Full Article
21 Likes
Arxiv
2d
281
Image Credit: Arxiv
Training-Free Dataset Pruning for Instance Segmentation
- Existing dataset pruning techniques primarily focus on classification tasks, limiting their applicability to more complex and practical tasks like instance segmentation.
- A novel Training-Free Dataset Pruning (TFDP) method is proposed for instance segmentation, addressing the challenges of pixel-level annotations, instance area variations, and class imbalances.
- The method leverages shape and class information from image annotations to design a Shape Complexity Score (SCS), refining it into Scale-Invariant (SI-SCS) and Class-Balanced (CB-SCS) versions.
- The proposed method achieves state-of-the-art results on VOC 2012, Cityscapes, and COCO datasets, and accelerates the pruning process by an average of 1349 times on COCO compared to adapted baselines.
Read Full Article
16 Likes
Arxiv
2d
0
Image Credit: Arxiv
Insights into dendritic growth mechanisms in batteries: A combined machine learning and computational study
- Researchers have conducted a study to understand dendritic growth mechanisms in batteries.
- A combined machine learning approach, along with computational methods, was used in the study.
- Two computer models, CNN-1 and CNN-2, were developed for dendrite growth prediction.
- The CNN-2 model, which integrates physical parameters, showed higher accuracy and sensitivity.
Read Full Article
Like
Arxiv
2d
118
Image Credit: Arxiv
Rewarding Graph Reasoning Process makes LLMs more Generalized Reasoners
- Developing advanced reasoning capabilities in Large Language Models (LLMs) is a challenge.
- Process Reward Models (PRMs) show promise in enhancing reasoning, particularly in mathematical reasoning.
- In this work, GraphSILO dataset is introduced for graph reasoning problems with step-wise labels.
- Experimental results show that GraphPRM significantly improves LLM performance in graph reasoning tasks and other domains.
Read Full Article
7 Likes
Arxiv
2d
384
Image Credit: Arxiv
Impact of Fasteners on the Radar Cross-Section performance of Radar Absorbing Air Intake Duct
- An aircraft's air intake duct contributes significantly to its radar cross-section performance.
- Radar absorbing materials (RAM) are used for RCS reduction in the air intake duct.
- The impact of rivets on the RCS performance of the air intake duct has not been studied extensively.
- This paper evaluates the RCS performance with different rivet configurations and surface area percentages.
Read Full Article
23 Likes
Arxiv
2d
111
Image Credit: Arxiv
PABBO: Preferential Amortized Black-Box Optimization
- Preferential Bayesian Optimization (PBO) is a method to learn latent user utilities from preferential feedback over designs.
- PBO relies on a statistical surrogate model, usually a Gaussian process, and an acquisition strategy to select the next candidate pair.
- A new approach called Preferential Amortized Black-Box Optimization (PABBO) fully amortizes PBO by meta-learning both the surrogate and acquisition function.
- PABBO outperforms Gaussian process-based strategies in terms of both computational speed and accuracy on synthetic and real-world datasets.
Read Full Article
6 Likes
Arxiv
2d
114
Image Credit: Arxiv
Explainable Classifier for Malignant Lymphoma Subtyping via Cell Graph and Image Fusion
- A novel explainable Multi-Instance Learning (MIL) framework is proposed for malignant lymphoma subtype classification.
- The framework integrates cell distribution characteristics and image information to identify subtype-specific Regions of Interest (ROIs).
- The proposed method achieves high-accuracy subtyping by fusing cell graph and image features using a Mixture-of-Experts (MoE) approach.
- Experiments demonstrate that the approach achieves state-of-the-art accuracy and provides region-level and cell-level explanations.
Read Full Article
6 Likes
Arxiv
2d
126
Image Credit: Arxiv
Revisiting CAD Model Generation by Learning Raster Sketch
- The integration of deep generative networks into generating Computer-Aided Design (CAD) models has garnered attention.
- A novel framework called RECAD is introduced for generating raster sketches and 3D extrusions in CAD models.
- RECAD breaks the limitations of traditional methods by representing sketches as raster images, providing enhanced geometric representation capabilities.
- Experimental results indicate that RECAD achieves strong performance in unconditional generation, conditional generation, and output editing.
Read Full Article
7 Likes
Arxiv
2d
198
Image Credit: Arxiv
Using Synthetic Images to Augment Small Medical Image Datasets
- Recent years have witnessed a growing interest in using deep learning for medical imaging.
- DL models require large labeled datasets, which is often a limitation in medical imaging.
- The study explores the use of synthetic images generated by a conditional variant of StyleGAN2 to augment small medical imaging datasets.
- The results indicate that the augmentation did not improve the segmentation performance, requiring further analysis.
Read Full Article
11 Likes
Arxiv
2d
348
Image Credit: Arxiv
Dialogue Without Limits: Constant-Sized KV Caches for Extended Responses in LLMs
- Autoregressive Transformers rely on Key-Value (KV) caching to accelerate inference, but the linear growth of the KV cache with context length leads to excessive memory consumption and bandwidth constraints.
- The proposed MorphKV technique maintains a constant-sized KV cache while preserving accuracy by adaptively ranking tokens through correlation-aware selection.
- MorphKV iteratively refines the KV cache via lightweight updates guided by attention patterns of recent tokens, capturing inter-token correlation with greater accuracy.
- Studies show 52.9% memory savings and 18.2% higher accuracy compared to prior works, making MorphKV suitable for efficient real-world deployment.
Read Full Article
20 Likes
Arxiv
2d
269
Image Credit: Arxiv
LLM-Fusion: A Novel Multimodal Fusion Model for Accelerated Material Discovery
- Discovering materials with desirable properties in an efficient way remains a significant problem in materials science.
- LLM-Fusion is a novel multimodal fusion model that integrates diverse representations for accurate material property prediction.
- The model leverages large language models (LLMs) to combine different sources of information, such as SMILES, SELFIES, text descriptions, and molecular fingerprints.
- LLM-Fusion outperforms traditional methods in terms of accuracy and shows promising results on multiple prediction tasks.
Read Full Article
16 Likes
Arxiv
2d
376
Image Credit: Arxiv
Powerful rank verification for multivariate Gaussian data with any covariance structure
- Researchers have developed a powerful rank verification method for multivariate Gaussian data with any covariance structure.
- The method leverages tools from selective inference to make inferences about the largest K means.
- It is shown that the developed procedure draws the desired inference when the two-sided difference-of-means test rejects.
- The method is considered inadmissible for simultaneous inference approaches when n > 2.
Read Full Article
22 Likes
Arxiv
2d
91
Image Credit: Arxiv
Vector Copula Variational Inference and Dependent Block Posterior Approximations
- Variational inference (VI) is a popular method to estimate statistical and econometric models.
- This paper proposes using vector copulas to capture dependence between the blocks parsimoniously.
- Tailored multivariate marginals are constructed using learnable cyclically monotone transformations.
- The approach demonstrates efficacy and versatility in producing more accurate posterior approximations than benchmark VI methods.
Read Full Article
5 Likes
Arxiv
2d
111
Image Credit: Arxiv
Learning Stochastic Dynamical Systems with Structured Noise
- Stochastic differential equations (SDEs) are used for modeling in various fields.
- A nonparametric framework is presented to learn drift and diffusion terms in SDE systems.
- The framework is effective in systems with singular covariance matrix and high dimensionality reduction.
- The framework is demonstrated through examples in physics, biology, and collective dynamics.
Read Full Article
6 Likes
Arxiv
2d
225
Image Credit: Arxiv
Identity documents recognition and detection using semantic segmentation with convolutional neural network
- Object recognition and detection are well-studied problems with a developed set of almost standard solutions.
- This paper proposes a new architecture based on an artificial convolutional neural network and semantic segmentation for the recognition and detection of identity documents.
- The research aims to evaluate the deep learning detection model trained on a mobile identity document video dataset, achieving an accuracy above 0.75 for the intersection over union (IoU) threshold value of 0.8.
- The study also verifies the feasibility of running the model on industrial one-board microcomputer or smartphone hardware.
Read Full Article
13 Likes
For uninterrupted reading, download the app