A naukri.com initiative
ML News
Arxiv
3h
230
Image Credit: Arxiv
Extract Free Dense Misalignment from CLIP
- Recent vision-language foundation models often produce misalignments in their outputs.
- A novel approach called CLIP4DM has been proposed to detect dense misalignments between image and text.
- CLIP4DM revamps the gradient-based attribution computation method to indicate misalignment.
- CLIP4DM demonstrates state-of-the-art performance and efficiency in detecting misalignments.
Read Full Article
13 Likes
Arxiv
3h
220
Image Credit: Arxiv
A Statistical Framework for Ranking LLM-Based Chatbots
- A statistical framework is proposed for ranking LLM-based chatbots.
- The framework enhances the ability to handle ties in pairwise comparisons.
- It models covariance between competitors for deeper performance insights.
- The framework demonstrates substantial improvements in modeling pairwise comparison data.
Read Full Article
13 Likes
Arxiv
3h
126
Image Credit: Arxiv
Discovery of 2D Materials via Symmetry-Constrained Diffusion Model
- A symmetry-constrained diffusion model (SCDM) has been introduced to generate 2D materials that adhere to symmetry principles.
- The model incorporates Wyckoff positions and space group symmetries to ensure the diversity and quality of generated structures.
- Through DFT calculations, 843 energetically stable materials were identified out of 2,000 candidate structures.
- Six selected candidates were further analyzed for stability, including phonon band structure and electronic properties evaluations.
Read Full Article
7 Likes
Arxiv
3h
260
Image Credit: Arxiv
Gaussian entropic optimal transport: Schr\"odinger bridges and the Sinkhorn algorithm
- Entropic optimal transport problems are important in machine learning and generative modeling.
- The Sinkhorn algorithm is commonly used to solve these problems in finite spaces.
- This article presents a finite-dimensional recursive formulation of the Sinkhorn algorithm for general Gaussian multivariate models.
- The algorithm is closely related to the Kalman filter and provides closed form expressions of entropic transport maps and Schrödinger bridges.
Read Full Article
15 Likes
Arxiv
3h
156
Image Credit: Arxiv
SoK: On the Offensive Potential of AI
- The offensive potential of AI is a growing concern in society, with evidence showing its use in violating security and privacy objectives.
- This paper aims to provide a systematic analysis of the heterogeneous capabilities of offensive AI, considering risks to humans and systems.
- The authors analyze 95 research papers, 38 InfoSec briefings, a user study with 549 individuals, and the opinions of 12 experts to reveal overlooked ways in which AI can be used offensively.
- The findings not only highlight current threats but also lay the groundwork for addressing this issue in the future.
Read Full Article
9 Likes
Arxiv
3h
23
Image Credit: Arxiv
VORTEX: A Spatial Computing Framework for Optimized Drone Telemetry Extraction from First-Person View Flight Data
- This paper presents the Visual Optical Recognition Telemetry EXtraction (VORTEX) system for extracting and analyzing drone telemetry data from First Person View (FPV) Uncrewed Aerial System (UAS) footage.
- VORTEX employs MMOCR, a PyTorch-based Optical Character Recognition (OCR) toolbox, to extract telemetry variables from drone Heads Up Display (HUD) recordings, utilizing advanced image preprocessing techniques.
- Results demonstrate that the 5-second sampling rate, utilizing 4.07% of available frames, provides the optimal balance with a point retention rate of 64% and mean speed accuracy within 4.2% of the 1-second baseline while reducing computational overhead by 80.5%.
- This research is the first of its kind, providing quantitative benchmarks for establishing a robust framework for drone telemetry extraction and analysis using open-source tools and spatial libraries.
Read Full Article
1 Like
Arxiv
3h
253
Image Credit: Arxiv
Subsampling, aligning, and averaging to find circular coordinates in recurrent time series
- Researchers have introduced a new algorithm for finding robust circular coordinates on recurrent time series data, such as neuronal recordings of C. elegans.
- The algorithm corrects for uneven sampling density by adapting the method of averaging coordinates in manifold learning.
- Rejection sampling is used to address inhomogeneous sampling, and Procrustes matching is applied to align and average the subsamples.
- The technique is validated on synthetic data sets and neuronal activity recordings, revealing a topological model for C. elegans' neuronal trajectories.
Read Full Article
15 Likes
Arxiv
3h
50
Image Credit: Arxiv
GCN-ABFT: Low-Cost Online Error Checking for Graph Convolutional Networks
- Graph convolutional networks (GCNs) are popular for building machine-learning applications for graph-structured data.
- This work introduces GCN-ABFT, a cost-effective approach for error detection in GCN accelerators.
- GCN-ABFT calculates a checksum for the entire three-matrix product in a single GCN layer.
- Experimental results show that GCN-ABFT reduces the number of operations needed for checksum computation while maintaining fault-detection accuracy.
Read Full Article
3 Likes
Arxiv
3h
270
Image Credit: Arxiv
Token-Budget-Aware LLM Reasoning
- Reasoning is critical for large language models (LLMs) to excel in a wide range of tasks.
- Current LLM reasoning processes are unnecessarily lengthy and incur high token usage costs.
- A token-budget-aware LLM reasoning framework is proposed to dynamically estimate and manage token budgets based on reasoning complexity.
- Experiments show that this method effectively reduces token costs in Chain-of-Thought (CoT) reasoning with minimal performance reduction.
Read Full Article
16 Likes
Arxiv
3h
60
Image Credit: Arxiv
HNCI: High-Dimensional Network Causal Inference
- The problem of evaluating the effectiveness of a treatment or policy commonly appears in causal inference applications under network interference.
- A new method called High-Dimensional Network Causal Inference (HNCI) is proposed in this paper.
- HNCI provides valid confidence intervals for the average direct treatment effect on the treated (ADET) and a confidence set for the neighborhood size for interference effect.
- The method leverages a linear regression formulation and existing literature from linear regression and homogeneity pursuit to conduct valid statistical inferences with theoretical guarantees.
Read Full Article
3 Likes
Arxiv
3h
313
Image Credit: Arxiv
Scalable Quantum-Inspired Optimization through Dynamic Qubit Compression
- Researchers have developed a quantum-inspired framework that compresses large Ising models to fit available quantum hardware.
- The framework utilizes a physics-inspired GNN architecture to capture complex interactions in Ising models and predict alignments among neighboring qubits.
- By progressively merging aligned qubits, the model size can be reduced while preserving the optimization structure.
- Numerical studies have shown that the method effectively reduces instance size without compromising solution quality on quantum annealers.
Read Full Article
18 Likes
Arxiv
3h
286
Image Credit: Arxiv
ReducedLUT: Table Decomposition with "Don't Care" Conditions
- Lookup tables (LUTs) are commonly used to store precomputed values for mathematical computations.
- ReducedLUT is a novel method to reduce the size of LUTs by injecting don't cares into the compression process.
- This method introduces more self-similarities in LUTs which can be exploited using known decomposition techniques.
- In a machine learning application, this method achieves up to $1.63 imes$ reduction in LUT utilization with minimal accuracy degradation.
Read Full Article
17 Likes
Arxiv
3h
206
Image Credit: Arxiv
Exploring Embedding Priors in Prompt-Tuning for Improved Interpretability and Control
- The study explores the role of embedding collapse in Prompt-Tuning for language models.
- Embedding priors are designed and compared with Soft and Deep Prompt-Tuning methods.
- The findings suggest that priors strongly influence the position of tuned embeddings.
- The research raises questions about the significance of a single activation cluster for large language models' generalization abilities.
Read Full Article
12 Likes
Arxiv
3h
200
Image Credit: Arxiv
Resolution-Robust 3D MRI Reconstruction with 2D Diffusion Priors: Diverse-Resolution Training Outperforms Interpolation
- Deep learning-based 3D imaging, especially MRI, faces challenges due to limited availability of 3D training data.
- Existing methods for 3D MRI reconstruction with 2D diffusion priors suffer from decreased performance when voxel size varies.
- Researchers propose a resolution-robust approach using diffusion-guided regularization of randomly sampled 2D slices.
- Model-based approaches fail to bridge the performance gap, while training the diffusion model on various resolutions provides a resolution-robust method without compromising accuracy.
Read Full Article
12 Likes
Arxiv
3h
166
Image Credit: Arxiv
Decentralized Intelligence in GameFi: Embodied AI Agents and the Convergence of DeFi and Virtual Ecosystems
- A new research paper introduces the concept of integrating advanced embodied AI agents into GameFi platforms.
- These AI agents, developed using cutting-edge large language models (LLMs), can provide proactive, adaptive, and contextually rich interactions with players.
- The integration of AI agents with blockchain technology establishes a consensus-driven, decentralized GameFi ecosystem.
- This approach enhances player immersion, retention, and economic participation, bridging traditional gaming with Web3 technologies.
Read Full Article
10 Likes
For uninterrupted reading, download the app