A naukri.com initiative
ML News
Arxiv
1w
228
Image Credit: Arxiv
The Importance of Being Discrete: Measuring the Impact of Discretization in End-to-End Differentially Private Synthetic Data
- Differentially Private (DP) generative marginal models are used to release synthetic tabular datasets while providing privacy guarantees.
- This study measures the impact of different discretization strategies on the utility of DP generative models.
- Optimizing the choice of discretizer and number of bins can improve utility by almost 30%.
- Applying DP during the discretization process can mitigate the vulnerability to membership inference attacks.
Read Full Article
13 Likes
Arxiv
1w
304
Image Credit: Arxiv
RayFronts: Open-Set Semantic Ray Frontiers for Online Scene Understanding and Exploration
- Open-set semantic mapping is crucial for open-world robots.
- RayFronts is a unified representation that enables both dense and beyond-range efficient semantic mapping.
- RayFronts provides improved zero-shot 3D semantic segmentation performance and enhanced throughput compared to traditional methods.
- RayFronts reduces search volume more efficiently than online baselines, making it a suitable option for online scene understanding and exploration.
Read Full Article
18 Likes
Arxiv
1w
315
Image Credit: Arxiv
Neural Signal Compression using RAMAN tinyML Accelerator for BCI Applications
- Researchers propose a neural signal compression scheme using Convolutional Autoencoders (CAEs) for high-quality, multi-channel neural recording.
- The compression ratio achieved is up to 150 for compressing local field potentials (LFPs).
- The CAE encoder section is implemented on RAMAN, an energy-efficient tinyML accelerator designed for edge computing, and deployed on an Efinix Ti60 FPGA.
- The compressed neural data from RAMAN is reconstructed offline, yielding superior signal-to-noise and distortion ratios (SNDR) as well as high R2 scores.
Read Full Article
19 Likes
Arxiv
1w
0
Image Credit: Arxiv
Using ML filters to help automated vulnerability repairs: when it helps and when it doesn't
- The acceptance of candidate patches in automated program repair can be based on testing oracles or ML models.
- ML models can quickly classify patches, allowing more candidate patches to be generated.
- However, when the model predictions are unreliable, they cannot replace the more reliable oracles based on testing.
- A proposed solution is to use an ML model as a preliminary filter before the traditional filter based on testing.
Read Full Article
Like
Arxiv
1w
384
Image Credit: Arxiv
$\Pi$-NeSy: A Possibilistic Neuro-Symbolic Approach
- The article introduces a neuro-symbolic approach that combines low-level perception with high-level reasoning tasks.
- The goal is to determine the degree of possibility that an input instance belongs to a target concept.
- Intermediate concepts are used for explanation purposes and to justify the classification of an input instance.
- Efficient methods for defining the matrix relation and the equation system in a possibilistic rule-based system are also presented.
Read Full Article
23 Likes
Arxiv
1w
93
Image Credit: Arxiv
Enhancing Downstream Analysis in Genome Sequencing: Species Classification While Basecalling
- A novel method is introduced for metagenomic profiling in genome sequencing.
- The method involves simultaneous basecalling and multi-class genome classification.
- State-of-the-art accuracy is achieved in both basecalling and classification.
- The method has implications for improving metagenomic profiling in future studies.
Read Full Article
5 Likes
Arxiv
1w
333
Image Credit: Arxiv
DeduCE: Deductive Consistency as a Framework to Evaluate LLM Reasoning
- Frontier large language models can struggle on high school math problems outside standard benchmarks.
- A deductive consistency metric is proposed to analyze chain-of-thought output from language models.
- The metric evaluates performance on understanding input premises and inferring conclusions over multiple reasoning hops.
- Language models are found to be robust to increasing number of input premises but suffer accuracy decay with increased reasoning hops.
Read Full Article
20 Likes
Arxiv
1w
46
Image Credit: Arxiv
AssistanceZero: Scalably Solving Assistance Games
- Assistance games are a promising alternative to reinforcement learning from human feedback for training AI assistants.
- A new approach called AssistanceZero is presented, which extends AlphaZero with a neural network to solve assistance games in complex environments.
- AssistanceZero outperforms model-free RL algorithms and imitation learning in a Minecraft-based assistance game.
- In a human study, the AssistanceZero-trained assistant significantly reduces the number of actions required to complete building tasks.
Read Full Article
2 Likes
Arxiv
1w
50
Image Credit: Arxiv
Are We Done with Object-Centric Learning?
- Object-centric learning (OCL) focuses on learning representations that encode isolated objects in a scene.
- Recent advancements in sample-efficient segmentation models allow the separation of objects in the pixel space.
- A new training-free probe called Object-Centric Classification with Applied Masks (OCCAM) outperforms slot-based OCL methods in out-of-distribution (OOD) generalization.
- Challenges in real-world applications and fundamental questions related to object perception in human cognition remain.
Read Full Article
3 Likes
Arxiv
1w
393
Image Credit: Arxiv
Fast Convex Optimization for Two-Layer ReLU Networks: Equivalent Model Classes and Cone Decompositions
- Researchers have developed fast algorithms and robust software for convex optimization of two-layer neural networks with ReLU activation functions.
- The work leverages a convex reformulation of the weight-decay penalized training problem as a set of group-ℓ₁-regularized data-local models, utilizing polyhedral cone constraints.
- In the case of zero-regularization, the problem is exactly equivalent to unconstrained optimization of a convex 'gated ReLU' network with non-singular gates.
- The developed approaches outperform standard training heuristics and commercial interior-point solvers in terms of speed and performance.
Read Full Article
23 Likes
Arxiv
1w
253
Image Credit: Arxiv
SIGMA: An Efficient Heterophilous Graph Neural Network with Fast Global Aggregation
- Graph neural networks suffer from performance loss in cases of heterophily, where neighboring nodes are dissimilar.
- Existing heterophilous GNNs have limitations in efficient global aggregation on large-scale graphs.
- The SIGMA model integrates SimRank for efficient global heterophilous GNN aggregation.
- SIGMA achieves state-of-the-art performance and 5x acceleration on the large-scale pokec dataset.
Read Full Article
15 Likes
Arxiv
1w
182
Image Credit: Arxiv
GBG++: A Fast and Stable Granular Ball Generation Method for Classification
- A fast and stable granular ball generation method (GBG++) for classification is proposed in this paper.
- GBG++ method improves the stability and efficiency of existing granular ball generation methods by using the attention mechanism.
- The proposed GBG++ method calculates the distances from the data-driven center to undivided samples, resulting in improved effectiveness, robustness, and efficiency.
- Experimental results show that the GBG++ method outperforms existing granular ball-based classifiers and classical machine learning classifiers on 24 public benchmark datasets.
Read Full Article
10 Likes
Arxiv
1w
198
Image Credit: Arxiv
Navigating Explanatory Multiverse Through Counterfactual Path Geometry
- Counterfactual explanations are the de facto standard for interpreting decisions of predictive models.
- Existing approaches neglect the multiplicity of counterfactual paths.
- The concept of explanatory multiverse is introduced to encompass all possible counterfactual journeys.
- An all-in-one metric called opportunity potential is proposed to quantify the spatial properties of counterfactual trajectories.
Read Full Article
11 Likes
Arxiv
1w
101
Image Credit: Arxiv
Estimation of embedding vectors in high dimensions
- Embeddings are a basic feature extraction step in many machine learning models, especially in natural language processing.
- A probability model is used to study the learning capability of embeddings, where the correlation of random variables is related to the similarity of the embeddings.
- The low-rank approximate message passing (AMP) method can be used to learn the embeddings.
- The theoretical findings are validated through simulations on synthetic and real text data.
Read Full Article
6 Likes
Arxiv
1w
139
Image Credit: Arxiv
TESSERACT: Eliminating Experimental Bias in Malware Classification across Space and Time (Extended Version)
- Machine learning (ML) plays a pivotal role in detecting malicious software.
- Inflated results in malware detection are due to spatial and temporal biases in experimental design.
- TESSERACT introduces constraints for fair experiment design and proposes a new metric, AUT, for classifier robustness.
- Performance enhancements are possible through periodic tuning and mitigation strategies.
Read Full Article
8 Likes
For uninterrupted reading, download the app