Research library

Useful ideas.
Original sources.

Short editorial introductions to research and technical references. These are source summaries, not claims of proprietary HanovaDevs research.

AI & Machine Learning

Original paper · arXivJune 2017

Attention Is All You Need

The original Transformer paper, proposing an attention-based architecture for sequence modelling.

Read editorial introduction

What the paper contributes

This paper introduced the Transformer architecture, using attention in place of recurrence and convolutions for its sequence-transduction model. It evaluates the approach on translation tasks.

How to read it

Separate the architecture described in the paper from later models and production systems built around related ideas. Its reported results belong to the experimental settings in the paper, not every application using attention.

Original source

This is a short editorial introduction. Follow the source link for the authors, experiments, methodology, and limitations.

Read original source ↗

AI & Machine Learning

Original paper · arXivMay 2020

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

A foundational paper combining retrieval with language generation for knowledge-intensive tasks.

Read editorial introduction

The central idea

The paper combines a retrieval component with a sequence-to-sequence model. Retrieved material provides a non-parametric source of information alongside the model’s learned parameters.

Practical interpretation

Retrieval does not by itself guarantee a correct answer. A deployed system needs to evaluate the material it retrieves, the answer it generates, and the access controls around the information.

Original source

Read the original paper for its architecture, task-specific evaluation, and reported experimental results.

Read original source ↗

AI & Machine Learning

Original paper · arXivJune 2021

LoRA: Low-Rank Adaptation of Large Language Models

A parameter-efficient approach to adapting pretrained language models.

Read editorial introduction

The approach

LoRA freezes pretrained model weights and introduces trainable low-rank matrices for adaptation. The paper compares this approach with other adaptation methods in its chosen experimental settings.

Choosing an adaptation strategy

The appropriate method depends on the model, task, data, hardware, and deployment constraints. A result reported in a research benchmark should not be presented as a guaranteed saving in a client application.

Original source

Use the original publication to inspect the method, experiments, and trade-offs.

Read original source ↗

UX & Design

W3C recommendationLiving reference

WCAG 2.2: A Reference for Accessible Interfaces

The W3C reference for assessing accessibility of web content and interaction.

Read editorial introduction

Why this reference matters

WCAG organizes accessibility requirements around content being perceivable, operable, understandable, and robust. The recommendation defines testable success criteria and levels of conformance.

Use it with real testing

Check keyboard use, screen-reader interpretation, contrast, form labels, focus behavior, and other criteria relevant to the page. An automated scan covers only part of an assessment.

Original source

The W3C specification provides the authoritative criteria. A project should state its testing scope before claiming conformance.

Read original source ↗

Web Development

Google · web.devLiving reference

Web Vitals: Understand Loading, Responsiveness and Stability

A practical starting point for evaluating a website’s performance experience.

Read editorial introduction

Measure an experience

Web Vitals describes signals for evaluating page experience. Core Web Vitals focus on loading, responsiveness, and visual stability.

Keep the conditions clear

Distinguish laboratory testing from measurements of real visitors. Devices, connections, page content, and third-party services affect results. A controlled score should be reported with its test conditions.

Original source

Use the official guide for current metric definitions and recommended measurement tools.

Read original source ↗