Key papers in Artificial Intelligence
Pofolia’s corpus holds 86 papers from the Artificial Intelligence subfield (1958–2026). The list below starts with the most cited.
Most cited
Ranked by citation count. Because citations accumulate over time, this list naturally leans towards work published a few years ago; for where the field is now, see “recently added”.
Fitting Linear Mixed-Effects Models Using <b>lme4</b>
Journal of Statistical Software · 2015 · Q1 · SJR 3.00 · FWCI 6077.98 · 86,393 citations · Open access
The lme4 package in R provides the `lmer` function to fit linear mixed-effects models using maximum likelihood or REML estimation.
Adam: A Method for Stochastic Optimization
UvA-DARE (University of Amsterdam) · 2014 · FWCI 1726.23 · 84,793 citations
Adam is a new, computationally efficient algorithm for gradient-based optimization of stochastic objective functions, using adaptive estimates of lower-order moments.
Inline Hardware KV-Cache Compression for Long-Context Transformer Inference: An Architectural Case for a Memory-Path Compression Engine
DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · 2023 · FWCI 7347.72 · 77,000 citations · Open access
This paper presents an AI/TP system that automatically proves a significant portion of Mizar theorems, achieving 60% in a hammer setting and 75% when guided by human-selected premises.
Scikit-learn: Machine Learning in Python
arXiv (Cornell University) · 2012 · 63,878 citations · Open access
Scikit-learn is a new Python module that integrates a wide range of machine learning algorithms for medium-scale problems, designed for non-specialists.
XGBoost
2016 · 49,653 citations · Open access
XGBoost is a novel, scalable tree boosting system designed for high performance on large datasets, achieving state-of-the-art results in machine learning challenges.
AI-Assisted Pipeline for Dynamic Generation of Trustworthy Health Supplement Content at Scale
DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · 2018 · FWCI 2503.48 · 45,742 citations
A new framework addresses the challenge of answering qualitative spatial questions by integrating a geoparser, a crisp reasoning system, and answer extraction.
SciPy 1.0: fundamental algorithms for scientific computing in Python
Nature Methods · 2020 · Q1 · SJR 15.00 · FWCI 2401.92 · 38,420 citations
SciPy 1.0 marks a significant milestone for the open-source Python library, solidifying its role as a de facto standard for scientific computing with over a decade of development and widespread adoption.
Dropout: a simple way to prevent neural networks from overfitting
2014 · FWCI 1844.83 · 34,283 citations
A new technique called 'dropout' randomly removes units from neural networks during training, preventing them from becoming too specialized and significantly reducing overfitting.
Glove: Global Vectors for Word Representation
2014 · 33,863 citations
A new global log-bilinear regression model, Glove, efficiently learns word vector representations by training on a word-word co-occurrence matrix, achieving 75% accuracy on a word analogy task.
Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation
2014 · 24,614 citations · Open access
This paper introduces an RNN encoder-decoder model that learns to represent phrases and translate them, outperforming phrase-based statistical machine translation (SMT) on a French-to-English task.
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
arXiv (Cornell University) · 2015 · 24,393 citations · Open access
Batch Normalization, a new technique that normalizes layer inputs within the model architecture, significantly accelerates deep neural network training by reducing internal covariate shift.
Array programming with NumPy
Nature · 2020 · Q1 · SJR 19.00 · FWCI 1249.42 · 22,658 citations · Open access
NumPy is the core Python library enabling powerful array programming for scientific data analysis across diverse fields like physics, astronomy, and finance.
Xception: Deep Learning with Depthwise Separable Convolutions
2017 · 19,087 citations
A new deep convolutional neural network architecture, Xception, replaces Inception modules with depthwise separable convolutions, achieving superior performance on large-scale image classification tasks.
Array programming with NumPy
TUScholarShare (Temple University) · 2020 · FWCI 267.66 · 18,809 citations · Open access
NumPy is the core Python library enabling powerful, compact, and expressive array programming for scientific data analysis across diverse fields.
Grey Wolf Optimizer
Advances in Engineering Software · 2014 · Q1 · SJR 1.00 · FWCI 352.66 · 18,589 citations
A new meta-heuristic algorithm, the Grey Wolf Optimizer (GWO), is introduced, mimicking the social hierarchy and hunting behaviors of grey wolves.
Efficient Estimation of Word Representations in Vector Space
arXiv (Cornell University) · 2013 · 18,154 citations · Open access
This paper introduces two new model architectures that significantly improve the accuracy and reduce the computational cost of creating word vector representations from massive datasets.
Distributed Representations of Words and Phrases and their Compositionality
arXiv (Cornell University) · 2013 · 18,083 citations · Open access
This paper introduces extensions to the Skip-gram model for learning word vector representations, significantly improving training speed and vector quality through subsampling frequent words and a new negative sampling method.
HISTORIAE, History of Socio-Cultural Transformation as Linguistic Data Science. A Humanities Use Case
DROPS (Schloss Dagstuhl – Leibniz Center for Informatics) · 2019 · FWCI 1423.45 · 17,374 citations
This paper introduces a novel method for automatically learning problem instance features directly from high-level representations using a transformer encoder, improving the selection of optimal model-solver combinations for combinatorial optimization.
Detecting Functionality-Specific Vulnerabilities via Retrieving Individual Functionality-Equivalent APIs in Open-Source Repositories
Dagstuhl Research Online Publication Server · 2025 · 16,189 citations
A new approach, APISS, effectively detects functionality-specific software vulnerabilities by retrieving equivalent APIs using API documentation and signatures, achieving 0.81 Top-1 Accuracy compared to baselines' 0.55.
Density Estimation for Statistics and Data Analysis
2018 · FWCI 329.90 · 15,980 citations
This book aims to bridge the gap between technical density estimation research and its practical applications in statistics, making the subject more accessible.
Recently added
RhizoDB: A Bounded Activation-Flow Architecture for Graph-Based Memory Systems
Zenodo (CERN European Organization for Nuclear Research) · 2026 · 914 citations
RhizoDB introduces a novel graph-based memory architecture for agentic systems, utilizing bounded activation flow and stability-modulated temporal decay.
A Survey of Large Language Models
Frontiers of Computer Science · 2026 · Q1 · SJR 1.00 · FWCI 934.90 · 1,475 citations · Open access
This survey systematically reviews recent advancements in Large Language Models (LLMs) across four key dimensions: pre-training, post-training adaptation, utilization strategies, and evaluation methods.
Affordance-Compiled Intelligence: Observable-Only Cognitive Impedance Matching for No-Meta LLM-Integrated Systems
arXiv (Cornell University) · 2026 · 3,049 citations · Open access
This paper introduces Cognitive Impedance Matching Theory (CIMT), a new framework for enhancing LLM system reliability by redesigning the surrounding observable world, rather than improving model weights. CIMT treats system capability as a world-side compilation problem, defining operational claims through explicit objects and ledgers.
Engineering Non-Linear Decay Dynamics: Pulse-Level Control and Software-Defined Qubit Rescue on Superconducting Processors
Zenodo (CERN European Organization for Nuclear Research) · 2025 · 788 citations
Researchers demonstrated a "Software-Defined Hardware" optimization technique on superconducting qubits, using "Instruction-Level Calibration Injection" to apply AC Stark drives and bypass compiler limits, thereby mitigating decoherence from material defects.
DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
Nature · 2025 · Q1 · SJR 19.00 · FWCI 947.64 · 853 citations · Open access
A new reinforcement learning framework, DeepSeek-R1, enables large language models (LLMs) to develop advanced reasoning abilities without human-annotated examples.
Add this field to your daily feed
Pick your interests and new work in your area arrives every day, summarised. Full summaries live in the app.
Open the app