A naukri.com initiative
ML News
Arxiv
6d
222
Image Credit: Arxiv
Planning and Learning in Risk-Aware Restless Multi-Arm Bandit Problem
- In this work, a generalized restless multi-arm bandit problem with risk-awareness is addressed.
- Indexability conditions for risk-aware objective are established and a solution based on Whittle index is provided.
- A Thompson sampling approach is proposed for the learning problem with unknown transition probabilities, achieving bounded regret.
- Numerical experiments illustrate the efficacy of the method in reducing risk exposure in various applications.
Read Full Article
13 Likes
Arxiv
6d
172
Image Credit: Arxiv
K-Means Clustering With Incomplete Data with the Use of Mahalanobis Distances
- Effectively applying the K-means algorithm to clustering tasks with incomplete features remains an important research area.
- Recent work has shown that unifying K-means clustering and imputation into one single objective function yields superior results.
- In this work, a unified K-means algorithm that incorporates Mahalanobis distances is proposed.
- Extensive experiments demonstrate that the proposed algorithm consistently outperforms existing approaches in handling incomplete data.
Read Full Article
10 Likes
Arxiv
6d
308
Image Credit: Arxiv
MrSteve: Instruction-Following Agents in Minecraft with What-Where-When Memory
- Significant advances have been made in developing general-purpose embodied AI in environments like Minecraft through the adoption of LLM-augmented hierarchical approaches.
- Low-level controllers frequently become performance bottlenecks due to repeated failures, which is primarily caused by the absence of an episodic memory system.
- MrSteve (Memory Recall Steve) is a novel low-level controller equipped with Place Event Memory (PEM) that captures what, where, and when information from episodes, addressing the limitations of previous models.
- The proposed Exploration Strategy and Memory-Augmented Task Solving Framework improves task-solving and exploration efficiency compared to existing methods.
Read Full Article
18 Likes
Arxiv
6d
292
Image Credit: Arxiv
Embedding Byzantine Fault Tolerance into Federated Learning via Consistency Scoring
- Federated learning (FL) enables training a shared model without transmitting private data to a central server.
- FL is vulnerable to Byzantine attacks from compromised edge devices, degrading model performance.
- The proposed plugin integrates into existing FL techniques to achieve Byzantine resilience.
- By generating virtual data samples and evaluating model consistency scores, compromised devices can be filtered out, maintaining the benefits of FL.
Read Full Article
17 Likes
Arxiv
6d
45
Image Credit: Arxiv
Space to Policy: Scalable Brick Kiln Detection and Automatic Compliance Monitoring with Geospatial Data
- Air pollution kills 7 million people annually.
- The brick kiln sector contributes to economic development but also accounts for 8-14% of air pollution in India.
- Researchers have developed a scalable machine-learning pipeline to detect and classify brick kilns using satellite imagery.
- Automated compliance analysis based on government policies has been performed, and the study emphasizes the need for inclusive policies balancing environmental sustainability and the livelihoods of workers.
Read Full Article
2 Likes
Arxiv
6d
74
Image Credit: Arxiv
Achieving Group Fairness through Independence in Predictive Process Monitoring
- Predictive process monitoring focuses on forecasting future states of ongoing process executions.
- This work addresses group fairness in predictive process monitoring by investigating independence and ensuring predictions are unaffected by sensitive group membership.
- The study explores independence through metrics such as demographic parity and threshold-independent distribution-based alternatives.
- The effectiveness of fairness metrics and composite loss functions is validated through a controlled experimental setup.
Read Full Article
4 Likes
Arxiv
6d
12
Image Credit: Arxiv
Efficient Fine-Tuning of Single-Cell Foundation Models Enables Zero-Shot Molecular Perturbation Prediction
- Efficient Fine-Tuning of Single-Cell Foundation Models Enables Zero-Shot Molecular Perturbation Prediction
- Researchers present a study on predicting transcriptional responses to novel drugs for biomedical research and drug discovery.
- They leverage single-cell foundation models pre-trained on a wide range of single cells, allowing molecular conditioning while preserving biological representation.
- The approach enables efficient fine-tuning and zero-shot generalization to unseen cell lines, demonstrating state-of-the-art results in model performance.
Read Full Article
Like
Arxiv
6d
20
Image Credit: Arxiv
Tokenphormer: Structure-aware Multi-token Graph Transformer for Node Classification
- Graph Neural Networks (GNNs) are widely used in graph data mining tasks.
- Traditional GNNs face limitations in the receptive field during message passing processes.
- A new model called Tokenphormer utilizes fine-grained token-based representation learning to capture local and structural information.
- Tokenphormer achieves state-of-the-art performance on node classification tasks.
Read Full Article
1 Like
Arxiv
6d
185
Image Credit: Arxiv
Real-Time Decision-Making for Digital Twin in Additive Manufacturing with Model Predictive Control using Time-Series Deep Neural Networks
- This paper presents a Model Predictive Control (MPC) framework using a deep neural network for real-time decision-making in additive manufacturing.
- The framework, named Time-Series Dense Encoder (TiDE), can predict future states within the prediction horizon in one shot, accelerating the MPC process.
- Using Directed Energy Deposition (DED) additive manufacturing as a case study, the MPC achieves precise temperature tracking and melt pool depth control.
- Compared to a PID controller, the MPC provides smoother laser power profiles with competitive or better melt pool temperature control performance.
Read Full Article
11 Likes
Arxiv
6d
78
Image Credit: Arxiv
Logarithmic Regret for Nonlinear Control
- Researchers have addressed the problem of learning to control an unknown nonlinear dynamical system through sequential interactions.
- They focus on achieving fast sequential learning in continuous control problems where the system dynamics depend smoothly on unknown parameters.
- The study demonstrates that fast sequential learning is attainable if the optimal control policy is persistently exciting.
- Additionally, they derive a regret bound which grows with the square root of the number of interactions for non-persistently exciting optimal policies.
Read Full Article
4 Likes
Arxiv
6d
24
Image Credit: Arxiv
No Equations Needed: Learning System Dynamics Without Relying on Closed-Form ODEs
- Data-driven modeling of dynamical systems is a crucial area of machine learning.
- A new approach called direct semantic modeling is proposed, which predicts the behavior of a dynamical system directly from data.
- This approach bypasses the need for complex post-hoc analysis and enhances the transparency and flexibility of the resulting models.
- The direct semantic modeling approach simplifies the modeling pipeline and allows for intuitive inductive biases and direct editing of the model's behavior.
Read Full Article
1 Like
Arxiv
6d
242
Image Credit: Arxiv
Wasserstein Gradient Flows for Moreau Envelopes of f-Divergences in Reproducing Kernel Hilbert Spaces
- Researchers propose a regularization approach for f-divergences using Moreau envelopes in reproducing kernel Hilbert spaces.
- The regularization is achieved by employing a squared maximum mean discrepancy (MMD) associated with a characteristic kernel K.
- The authors analyze the gradients of MMD-regularized f-divergences and study Wasserstein gradient flows based on their findings.
- Proof-of-concept numerical examples are provided for flows starting from empirical measures.
Read Full Article
14 Likes
Arxiv
6d
152
Image Credit: Arxiv
Optimal Rates and Saturation for Noiseless Kernel Ridge Regression
- Kernel ridge regression (KRR) is a fundamental method for learning functions from finite samples.
- A comprehensive study of KRR in the noiseless regime reveals optimal convergence rates determined by eigenvalue decay and target function's smoothness.
- KRR exhibits extra-smoothness compared to typical functions in the native reproducing kernel Hilbert space (RKHS).
- A novel error bound for noisy KRR achieves minimax optimality in both noiseless and noisy regimes.
Read Full Article
9 Likes
Arxiv
6d
176
Image Credit: Arxiv
Understanding Optimal Feature Transfer via a Fine-Grained Bias-Variance Analysis
- Researchers have conducted a study on transfer learning to optimize downstream performance.
- They introduced a simple linear model that utilizes a pretrained feature transform.
- The researchers derived the exact asymptotics of the downstream risk and its fine-grained bias-variance decomposition.
- The study revealed that the optimal featurization is naturally sparse and undergoes a phase transition from hard selection to soft selection of relevant features.
Read Full Article
10 Likes
Arxiv
6d
314
Image Credit: Arxiv
Faster Linear Systems and Matrix Norm Approximation via Multi-level Sketched Preconditioning
- Researchers have developed a new class of preconditioned iterative methods for solving linear systems, resulting in faster runtimes for various linear algebraic problems.
- The methods involve constructing a low-rank Nyström approximation to the matrix using sparse random matrix sketching.
- The approximation is then used to create a preconditioner, which is inverted quickly using additional levels of random sketching and preconditioning.
- The research proves the convergence of the methods is dependent on the average condition number of the matrix, which improves as the rank of the Nyström approximation increases.
Read Full Article
18 Likes
For uninterrupted reading, download the app