Machine Learning (ML) Latest News and Trending articles from all top sources only on Techminis

A naukri.com initiative

New

Home

ML News

Arxiv

Image Credit: Arxiv

An Approach to Technical AGI Safety and Security

Artificial General Intelligence (AGI) presents transformative benefits and risks.
The approach focuses on addressing two areas of risk: misuse and misalignment.
To prevent misuse, the strategy includes identifying dangerous capabilities and implementing security measures.
To address misalignment, model-level mitigations and system-level security measures are proposed.

Read Full Article

1 Like

Arxiv

209

Image Credit: Arxiv

Corner-Grasp: Multi-Action Grasp Detection and Active Gripper Adaptation for Grasping in Cluttered Environments

Robotic grasping is a critical capability for robots to interact with their environment.
Researchers propose a method for effective grasping in cluttered bin-picking environments.
They utilize a multi-functional gripper combining suction and finger grasping.
The proposed method prevents collisions and successfully grasps objects in cluttered scenarios.

Read Full Article

12 Likes

Arxiv

Image Credit: Arxiv

CoRAG: Collaborative Retrieval-Augmented Generation

CoRAG is a framework that extends Retrieval-Augmented Generation (RAG) models to collaborative settings.
CoRAG allows clients to jointly train a shared model using a collaborative passage store.
CoRAG outperforms parametric collaborative learning methods and locally trained RAG models in low-resource scenarios.
The trade-off between leveraging a collectively enriched knowledge base and the potential risk of incorporating detrimental passages is a key consideration in collaborative RAG.

Read Full Article

5 Likes

Arxiv

125

Image Credit: Arxiv

Graphically Speaking: Unmasking Abuse in Social Media with Conversation Insights

Detecting abusive language in social media conversations is challenging due to the contextual nature of abusiveness.
Traditional Abusive Language Detection (ALD) models often overlook the conversational context, leading to unreliable performance metrics.
A novel approach is proposed in this paper using graph neural networks (GNNs) to model social media conversations as graphs, capturing comment relationships.
The GNN model outperforms context-agnostic baselines and linear context-aware methods, achieving significant improvements in F1 scores.

Read Full Article

7 Likes

Discover more

Arxiv

236

Image Credit: Arxiv

Representing Flow Fields with Divergence-Free Kernels for Reconstruction

Accurately reconstructing continuous flow fields from sparse or indirect measurements remains an open challenge.
A novel flow field reconstruction framework based on divergence-free kernels (DFKs) is introduced.
DFKs-Wen4 (matrix-valued radial basis functions derived from Wendland's C^4 polynomial) are identified as the optimal form of analytically divergence-free approximation for velocity fields.
Experiments demonstrate that DFKs-Wen4 outperform other divergence-free representations in reconstruction accuracy and computational efficiency.

Read Full Article

14 Likes

Arxiv

Image Credit: Arxiv

Gen-C: Populating Virtual Worlds with Generative Crowds

Researchers introduce Gen-C, a generative model for authoring high-level crowd behaviors in virtual environments.
Gen-C leverages a large language model to generate crowd scenarios which are expanded and generalized through simulations.
The method employs Variational Graph Auto-Encoders to learn graph structures and node features, enabling flexible generation of dynamic crowd interactions.
Gen-C showcases its potential for populating diverse virtual environments with agents exhibiting varied and dynamic behaviors.

Read Full Article

Arxiv

Image Credit: Arxiv

Is the Reversal Curse a Binding Problem? Uncovering Limitations of Transformers from a Basic Generalization Failure

Despite their impressive capabilities, LLMs exhibit a basic generalization failure known as the Reversal Curse.
The Reversal Curse in LLMs is attributed to the long-standing binding problem in cognitive science, neuroscience, and AI.
Transformers' limitations in conceptual binding cause the inconsistency and entanglements of concept representations, leading to the Reversal Curse.
A model design based on JEPA (Joint-Embedding Predictive Architecture) breaks the Reversal Curse and improves generalization by incorporating memory layers supporting disentangled concept representations.

Read Full Article

Arxiv

104

Image Credit: Arxiv

Hessian-aware Training for Enhancing DNNs Resilience to Parameter Corruptions

Deep neural networks are not resilient to parameter corruptions: even a single-bitwise error in their parameters in memory can cause an accuracy drop of over 10%.
Hessian-aware training is proposed as an approach to improve resilience to bitwise corruptions in neural network parameters.
The approach promotes models with flatter loss surfaces and shows a reduction in the number of bits leading to a significant accuracy drop.
This method can work synergistically with existing hardware and system-level defenses.

Read Full Article

6 Likes

Arxiv

320

Image Credit: Arxiv

Deep Representation Learning for Unsupervised Clustering of Myocardial Fiber Trajectories in Cardiac Diffusion Tensor Imaging

Researchers have developed a deep learning framework for unsupervised clustering of myocardial fibers in cardiac diffusion tensor imaging (DTI) data.
The framework combines a Bidirectional Long Short-Term Memory (LSTM) network to capture local sequential information along fibers, with a Transformer autoencoder to learn global shape features and incorporate anatomical context.
By clustering the learned representations using a density-based algorithm, the framework successfully identifies 33 to 62 robust clusters, capturing subtle differences in fiber trajectories.
This approach has the potential to improve surgical planning, characterize disease-related remodeling, and advance personalized cardiac care.

Read Full Article

19 Likes

Arxiv

Image Credit: Arxiv

Toward Interpretable Sleep Stage Classification Using Cross-Modal Transformers

Accurate sleep stage classification is significant for sleep health assessment.
A new cross-modal transformer-based method for sleep stage classification is proposed.
The method outperforms the state-of-the-art methods and eliminates the black-box behavior of deep-learning models.
Considerable reductions in the number of parameters and training time are achieved compared to the state-of-the-art methods.

Read Full Article

1 Like

Arxiv

167

Image Credit: Arxiv

Latent Covariate Shift: Unlocking Partial Identifiability for Multi-Source Domain Adaptation

Multi-source domain adaptation (MSDA) aims to learn a label prediction function for an unlabeled target domain by leveraging labeled data from multiple source domains and unlabeled data from the target domain.
Conventional MSDA approaches rely on covariate shift or conditional shift paradigms, assuming a consistent label distribution across domains. However, this limits their applicability in real-world scenarios where label distributions vary across domains.
To address this limitation, a new paradigm called latent covariate shift (LCS) is proposed, introducing greater variability and adaptability across domains. It allows for recovering the latent cause of the label variable, referred to as the latent content variable.
The proposed MSDA method based on LCS achieves exceptional performance on both simulated and real-world datasets by learning the label distribution conditioned on the identifiable latent content variable, accommodating substantial distribution shifts.

Read Full Article

10 Likes

Arxiv

334

Image Credit: Arxiv

Epistemic Monte Carlo Tree Search

The AlphaZero/MuZero (A/MZ) family of algorithms utilizes Monte Carlo Tree Search (MCTS) and learned models for remarkable success in various domains.
Epistemic MCTS (EMCTS) is introduced to address the uncertainty caused by learned models and enhance exploration in sparse reward environments.
When applied to the task of writing code in the Assembly language subleq, AZ with EMCTS achieves higher sample efficiency compared to the baseline AZ.
Search with EMCTS significantly outperforms equivalent methods without search for uncertainty estimation in solving hard-exploration benchmark Deep Sea, showcasing the benefits of search for epistemic uncertainty estimation.

Read Full Article

20 Likes

Arxiv

146

Image Credit: Arxiv

Provable Guarantees for Nonlinear Feature Learning in Three-Layer Neural Networks

Provable Guarantees for Nonlinear Feature Learning in Three-Layer Neural Networks
Deep learning theory seeks to understand how neural networks learn hierarchical features.
This study focuses on three-layer neural networks and their richer feature learning capabilities.
They present a theorem that bounds sample complexity and width needed for low test error when the target has hierarchical structure.

Read Full Article

8 Likes

Arxiv

111

Image Credit: Arxiv

Data-Driven Knowledge Transfer in Batch $Q^*$ Learning

In data-driven decision-making, knowledge transfer can help address data scarcity in new ventures.
The authors propose a framework of Transferred Fitted $Q$-Iteration algorithm for knowledge transfer.
The framework enables direct estimation of the optimal action-state function using both target and source data.
The approach shows improved learning error rates compared to single task learning, both theoretically and empirically.

Read Full Article

6 Likes

Arxiv

292

Image Credit: Arxiv

Learning Actionable Counterfactual Explanations in Large State Spaces

Recourse generators provide actionable insights, often through feature-based counterfactual explanations (CFEs).
Introducing three novel recourse types grounded in real-world actions: high-level continuous (hl-continuous), high-level discrete (hl-discrete), and high-level ID (hl-id) CFEs.
Proposing data-driven CFE generation approaches that quickly provide optimal CFEs for new agents.
Empirical evaluation shows the effectiveness of the proposed forms of recourse over low-level CFEs.

Read Full Article

17 Likes

For uninterrupted reading, download the app