A naukri.com initiative
ML News
Arxiv
3d
20
Image Credit: Arxiv
HiPART: Hierarchical Pose AutoRegressive Transformer for Occluded 3D Human Pose Estimation
- Existing 2D-to-3D human pose estimation methods struggle with occlusion issue due to limited input representation.
- Hierarchical Pose AutoRegressive Transformer (HiPART) is proposed to address the occlusion issue in 2D-to-3D lifting.
- HiPART generates hierarchical 2D dense poses from sparse 2D pose using a two-stage generative densification method.
- HiPART achieves state-of-the-art performance on single-frame-based 3D human pose estimation by improving robustness in occluded scenarios.
Read Full Article
1 Like
Arxiv
3d
0
Image Credit: Arxiv
Large Language Models Are Better Logical Fallacy Reasoners with Counterargument, Explanation, and Goal-Aware Prompt Formulation
- Large Language Models (LLMs) have improved our ability to process complex language, but detecting logical fallacies remains a challenge.
- A study introduces a novel prompt formulation approach for logical fallacy detection.
- The approach incorporates counterarguments, explanations, and goals to enrich the input text.
- The method shows substantial improvements over previous models in both supervised and unsupervised settings.
Read Full Article
Like
Arxiv
3d
148
Image Credit: Arxiv
KernelDNA: Dynamic Kernel Sharing via Decoupled Naive Adapters
- Dynamic convolution enhances model capacity by adaptively combining multiple kernels.
- KernelDNA is a lightweight convolution kernel plug-in that enables dynamic kernel specialization without altering the standard convolution structure.
- KernelDNA achieves state-of-the-art accuracy-efficiency balance among dynamic convolution variants.
- Codes for KernelDNA are available at https://github.com/haiduo/KernelDNA.
Read Full Article
8 Likes
Arxiv
3d
268
Image Credit: Arxiv
COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation
- Researchers propose COSMIC (Clique-Oriented Semantic Multi-space Integration for CLIP), a test-time adaptation framework for vision-language models (VLMs).
- COSMIC enhances adaptability through multi-granular, cross-modal semantic caching and graph-based querying mechanisms.
- The framework introduces Dual Semantics Graph (DSG) to capture rich semantic relationships by incorporating textual features, coarse-grained CLIP features, and fine-grained DINOv2 features.
- The Clique Guided Hyper-class component leverages structured class relationships to enhance prediction robustness in COSMIC.
Read Full Article
16 Likes
Arxiv
3d
260
Image Credit: Arxiv
DGSAM: Domain Generalization via Individual Sharpness-Aware Minimization
- A new DG algorithm, Decreased-overhead Gradual Sharpness-Aware Minimization (DGSAM), has been introduced.
- DGSAM aims to reduce sharpness consistently across domains while maintaining computational efficiency.
- Experiments show that DGSAM outperforms state-of-the-art DG methods in terms of robustness and performance.
- DGSAM reduces computational overhead compared to the popular approach of Sharpness-Aware Minimization (SAM).
Read Full Article
15 Likes
Arxiv
3d
296
Image Credit: Arxiv
Speculative End-Turn Detector for Efficient Speech Chatbot Assistant
- Spoken dialogue systems struggle with end-turn detection (ETD) - the ability to distinguish between user turn completion and hesitation.
- The ETD Dataset is introduced, which is the first public dataset for end-turn detection.
- A collaborative inference framework, SpeculativeETD, is proposed to improve real-time ETD in resource-constrained environments.
- SpeculativeETD significantly improves ETD accuracy while minimizing computation requirements.
Read Full Article
17 Likes
Arxiv
3d
16
Image Credit: Arxiv
Semantic-Preserving Transformations as Mutation Operators: A Study on Their Effectiveness in Defect Detection
- Recent advances in defect detection use language models.
- A study was conducted to determine the effectiveness of semantic-preserving transformations in improving defect detection tools.
- 28 publications with 94 different transformations were analyzed.
- Reusing shared semantic-preserving transformations proved to be challenging and did not improve the accuracy of defect detection models.
Read Full Article
Like
Arxiv
3d
240
Image Credit: Arxiv
Accelerated Stein Variational Gradient Flow
- Accelerated Stein Variational Gradient Flow is a new method for sampling from a target distribution.
- It is an accelerated version of the existing Stein variational gradient descent (SVGD) method.
- ASVGD is designed to be fast and efficient in high-dimensional sampling.
- Numerical examples show the effectiveness of ASVGD compared to SVGD and other sampling methods.
Read Full Article
14 Likes
Arxiv
3d
257
Image Credit: Arxiv
Codehacks: A Dataset of Adversarial Tests for Competitive Programming Problems Obtained from Codeforces
- Codehacks is a dataset of programming problems obtained from the Codeforces online judge platform.
- The dataset includes 288,617 error-inducing test cases referred to as 'hacks' for 5,578 programming problems.
- Each problem in the dataset is accompanied by a natural language description and the source code for 2,196 submitted solutions.
- The dataset aims to support data-driven creation of test suites, particularly for testing software synthesized from large language models.
Read Full Article
15 Likes
Arxiv
3d
108
Image Credit: Arxiv
Benchmarking Systematic Relational Reasoning with Large Language and Reasoning Models
- Large Language Models (LLMs) struggle with systematic reasoning, even when performing well on certain tasks.
- Post-training strategies based on reinforcement learning and chain-of-thought prompting have been seen as an improvement.
- Little is known about the potential of Large Reasoning Models(LRMs) beyond mathematics and programming.
- LLMs and LRMs still overall perform poorly, albeit better than random chance.
Read Full Article
6 Likes
Arxiv
3d
208
Image Credit: Arxiv
POINT$^{2}$: A Polymer Informatics Training and Testing Database
- The integration of machine learning (ML) techniques has propelled the advancement of polymer informatics in predicting polymer properties and discovering high-performance materials.
- However, the field lacks a standardized workflow that encompasses prediction accuracy, uncertainty quantification, ML interpretability, and polymer synthesizability.
- To address these challenges, a comprehensive benchmark database and protocol called POINT$^{2}$ (POlymer INformatics Training and Testing) has been introduced.
- The POINT$^{2}$ database provides a collection of ML models and polymer representations to achieve property predictions, uncertainty estimations, model interpretability, and template-based polymerization synthesizability.
Read Full Article
12 Likes
Arxiv
3d
140
Image Credit: Arxiv
Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model
- DFI-OmniStereo is a novel omnidirectional stereo matching method that leverages a pre-trained depth foundation model for accurate depth estimation.
- Omnidirectional depth perception is essential for mobile robotics applications, and camera-based setups provide a cost-effective solution.
- Existing omnidirectional stereo matching approaches face limitations in accuracy due to the lack of real-world data.
- DFI-OmniStereo achieves state-of-the-art results on the real-world Helvipad dataset, significantly reducing disparity mean absolute error (MAE) compared to previous methods.
Read Full Article
8 Likes
Arxiv
3d
392
Image Credit: Arxiv
Question-Aware Knowledge Graph Prompting for Enhancing Large Language Models
- Large Language Models (LLMs) struggle with tasks requiring external knowledge.
- Knowledge Graphs (KGs) can enhance reasoning, but existing methods demand costly fine-tuning or retrieve noisy KG information.
- Question-Aware Knowledge Graph Prompting (QAP) dynamically assesses KG relevance and incorporates question embeddings into reasoning.
- Experimental results demonstrate that QAP outperforms state-of-the-art methods in Multiple Choice Question Answering tasks.
Read Full Article
23 Likes
Arxiv
3d
352
Image Credit: Arxiv
Addressing Model Overcomplexity in Drug-Drug Interaction Prediction With Molecular Fingerprints
- Accurately predicting drug-drug interactions (DDIs) is crucial for pharmaceutical research and clinical safety.
- This study proposes a simpler approach using molecular representations like Morgan fingerprints, graph-based embeddings, and transformer-derived embeddings.
- The combination of these representations achieves competitive performance in DDI prediction tasks.
- The study highlights the importance of dataset curation and progressive complexity scaling in drug interaction prediction models.
Read Full Article
21 Likes
Arxiv
3d
212
Image Credit: Arxiv
DASH: Detection and Assessment of Systematic Hallucinations of VLMs
- DASH (Detection and Assessment of Systematic Hallucinations) is an automatic, large-scale pipeline designed to identify systematic hallucinations of Vision-Language Models (VLMs) on real-world images in an open-world setting.
- The pipeline utilizes DASH-OPT for image-based retrieval, optimizing over the 'natural image manifold' to generate images that mislead the VLM and expose its object hallucinations.
- Applying DASH to PaliGemma and two LLaVA-NeXT models, it identifies more than 19k clusters with 950k images where the VLM hallucinates an object across 380 object classes.
- The study also demonstrates that fine-tuning PaliGemma with the model-specific images obtained using DASH mitigates object hallucinations.
Read Full Article
12 Likes
For uninterrupted reading, download the app