Highlight

BEAST 2.5: An advanced software platform for Bayesian evolutionary analysis

Describes new tools and models in the BEAST 2.5 platform for Bayesian phylogenetic, population genetic, and phylodynamic inference.

Bayesian phylogenetic inference increasingly combines evidence from many independent sources, such as genome sequences, sampling dates, phenotypes, fossil occurrences, and biogeographic ranges, into a single joint model that is conceptually and computationally challenging. Software platforms that let researchers compose compatible sub-models into a full model hierarchy are central to this. This paper describes major new developments in the BEAST 2 platform, culminating in the 2.5 release, which expands joint inference across data types, non-tree models, and phylodynamics.

Based on: BEAST 2.5: An advanced software platform for Bayesian evolutionary analysis · bioRxiv

Curated by Aramai Editorial

Read summary →
Highlight

A trust-based consumer decision-making model in electronic commerce: The role of trust, perceived risk, and their antecedents

Develops and tests a trust-based model of how online consumers' trust and perceived risk shape their e-commerce purchasing decisions.

This study asks whether trust and risk matter in consumers' e-commerce purchasing decisions, what their antecedents are, and how they affect purchase intent. The authors develop a theoretical framework of the trust-based decision process and test it via structural equation modeling on Internet purchasing data from a web survey. Results show trust and perceived risk strongly influence purchasing decisions. Disposition to trust, reputation, privacy and security concerns, website information quality, and company reputation strongly affect trust, while a third-party seal did not.

Based on: A trust-based consumer decision-making model in electronic commerce: The role of trust, perceived risk, and their antecedents · Decision Support Systems

Curated by Aramai Editorial

Read summary →
Highlight

Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context

Introduces the Gemini 1.5 model family, compute-efficient multimodal models that recall and reason over context up to millions of tokens.

This report introduces the Gemini 1.5 family of compute-efficient multimodal models that recall and reason over fine-grained information from millions of tokens of context, including long documents and hours of video and audio. It includes an updated Gemini 1.5 Pro and a lighter, efficient Gemini 1.5 Flash. The models achieve near-perfect long-context retrieval, advance long-document and long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra. Recall exceeds 99% up to at least 10M tokens, and given a manual the model learns Kalamang translation.

Based on: Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context · arXiv.org

Curated by Aramai Editorial

Read summary →
Highlight

Let's Verify Step by Step

Shows that process supervision outperforms outcome supervision for training reliable LLM reasoning, and releases the PRM800K dataset.

Large language models can perform complex multi-step reasoning but still make logical mistakes. This work compares two ways to train more reliable models: outcome supervision, giving feedback on the final result, and process supervision, giving feedback on each intermediate step. The authors find process supervision significantly outperforms outcome supervision on the challenging MATH dataset, with their model solving 78% of a representative test subset. They show active learning improves process supervision, and release PRM800K, 800,000 step-level human feedback labels.

Based on: Let's Verify Step by Step · International Conference on Learning Representations

Curated by Aramai Editorial

Read summary →
Highlight

The HHpred interactive server for protein homology detection and structure prediction

HHpred is a fast web server for remote protein homology detection and structure prediction using pairwise profile HMM comparison.

HHpred is a fast server for remote protein homology detection and structure prediction, and the first to implement pairwise comparison of profile hidden Markov models. It can search databases such as PDB, SCOP, Pfam, SMART, COGs, and CDD, accepting either a single query sequence or a multiple alignment as input, and returns results within minutes in a PSI-BLAST-like format. Options include local or global alignment and secondary-structure scoring. HHpred produces query-template alignments and 3D structural models built by MODELLER, demonstrated by analyzing SpoVT from Bacillus subtilis.

Based on: The HHpred interactive server for protein homology detection and structure prediction · Nucleic Acids Res.

Curated by Aramai Editorial

Read summary →
Highlight

Generative Adversarial Networks: An Overview

An overview of generative adversarial networks for the signal processing community, covering methods, applications, and open challenges.

This review introduces generative adversarial networks (GANs), which learn deep representations without extensively annotated data by deriving backpropagation signals through a competitive process between a pair of networks. The learned representations support applications including image synthesis, semantic image editing, style transfer, super-resolution, and classification. Aimed at the signal processing community, the article draws on familiar analogies, identifies methods for training and constructing GANs, and points to remaining theoretical and practical challenges.

Based on: Generative Adversarial Networks: An Overview · IEEE Signal Processing Magazine

Curated by Aramai Editorial

Read summary →
Highlight

Image Segmentation Using Deep Learning: A Survey

A comprehensive survey of deep learning approaches to semantic and instance image segmentation, their datasets, and performance.

This survey reviews recent deep-learning approaches to image segmentation, a core computer vision task underpinning scene understanding, medical imaging, surveillance, and augmented reality. It covers pioneering semantic and instance segmentation methods, including convolutional pixel-labeling networks, encoder-decoder architectures, multiscale and pyramid approaches, recurrent networks, attention models, and adversarial generative models. The authors analyze their strengths and challenges, review widely used datasets, compare performances, and discuss future directions.

Based on: Image Segmentation Using Deep Learning: A Survey · IEEE Transactions on Pattern Analysis and Machine Intelligence

Curated by Aramai Editorial

Read summary →
Highlight

Reinforcement learning in robotics: A survey

A survey of reinforcement learning for generating robot behaviors, examining key challenges, successes, and algorithmic design choices.

This article surveys reinforcement learning as a framework for designing sophisticated, hard-to-engineer robot behaviors, arguing the two fields relate much as physics does to mathematics. It highlights central challenges of robot RL alongside notable successes, and studies how algorithms, representations, and prior knowledge tamed the domain's complexity. A particular focus is the choice between model-based and model-free methods and between value-function and policy-search approaches. Through a detailed example, the authors show how RL applies and note open questions.

Based on: Reinforcement learning in robotics: A survey · Int. J. Robotics Res.

Curated by Aramai Editorial

Read summary →
Highlight

StarGAN: Unified Generative Adversarial Networks for Multi-domain Image-to-Image Translation

Introduces StarGAN, a single scalable GAN model performing image-to-image translation across multiple domains.

Image-to-image translation has succeeded for two domains, but existing approaches scale poorly because a separate model must be built for every pair of domains. The paper proposes StarGAN, a scalable approach performing translations across multiple domains using only a single model. Its unified architecture allows simultaneous training on multiple datasets with different domains within one network, yielding superior image quality and flexible translation of an input to any target domain. Effectiveness is shown on facial attribute transfer and facial expression synthesis.

Based on: StarGAN: Unified Generative Adversarial Networks for Multi-domain Image-to-Image Translation · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition

Curated by Aramai Editorial

Read summary →
Highlight

Simplifying Graph Convolutional Networks

Simplifies GCNs by removing nonlinearities and collapsing weight matrices, yielding a linear low-pass-filter model with comparable accuracy.

Graph Convolutional Networks became the de facto method for learning graph representations but inherited unnecessary complexity from deep learning. The paper reduces this by successively removing nonlinearities and collapsing weight matrices between consecutive layers. The resulting linear model corresponds to a fixed low-pass filter followed by a linear classifier. These simplifications do not hurt accuracy on many downstream tasks, while the model scales to larger datasets, is interpretable, and gives up to two orders of magnitude speedup over FastGCN.

Based on: Simplifying Graph Convolutional Networks · International Conference on Machine Learning

Curated by Aramai Editorial

Read summary →
Highlight

Residual Dense Network for Image Super-Resolution

Proposes the Residual Dense Network, which fully exploits hierarchical features from all conv layers for image super-resolution.

Very deep CNNs have succeeded at image super-resolution but most fail to fully use hierarchical features from low-resolution inputs, limiting performance. The paper proposes the Residual Dense Network (RDN), exploiting features from all convolutional layers. Its residual dense block extracts abundant local features via densely connected layers with direct connections from preceding blocks, forming a contiguous memory mechanism. Local and global feature fusion then adaptively combine local and holistic features, and RDN performs favorably against state-of-the-art methods.

Based on: Residual Dense Network for Image Super-Resolution · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition

Curated by Aramai Editorial

Read summary →
Highlight

Optimization Methods for Large-Scale Machine Learning

Reviews numerical optimization for machine learning, centering on stochastic gradient methods and directions for next-generation algorithms.

This review examines numerical optimization algorithms for machine learning. Through case studies on text classification and deep neural network training, it discusses how optimization problems arise and why they are challenging. A central theme is that large-scale machine learning is a setting where the stochastic gradient (SG) method has been central while conventional gradient-based techniques falter. The authors present a theory of a versatile SG algorithm and highlight next-generation directions, including noise-reduction and second-order methods.

Based on: Optimization Methods for Large-Scale Machine Learning · SIAM Review

Curated by Aramai Editorial

Read summary →