ngi-search
12 de febrero de 2024
The NGI Search project, co-funded by the European Union's Horizon 2020 research and innovation program, has unveiled the outcomes of its third Open Call.
This initiative aims to assist diverse beneficiaries, including researchers, developers, internet activists, hackers, SMEs, and startups, in adopting and developing solutions for data and resource discovery on the Internet.
Ninety-four proposals were submitted for this call, resulting in the selection of ten proposals that have sought financial support. These beneficiaries will receive technical and business support, from Open Source Licensing to Market Readiness.
👉 More information here.
✨Meet the selected projects✨
✅OKLLM: Online Knowledge Search for LLM innovations
Nowadays, a breakthrough era in online content discovery and search is being ushered in by the increased availability of LLMs. However, LLMs are resource and computationally intensive. Additionally, AI-generated material can occasionally contain bias or false information.
In this research, the project seeks to fill these gaps through knowledge distillation, knowledge graph generation and verification, bias detection, and transfer learning. It will be made up of three components.
📌 The first component will employ knowledge distillation to handle bias detection tasks to enhance search results and lessen computationally taxing tasks.
📌 The second component will use the knowledge graph to solve the hallucination phenomenon and improve search results.
📌 The third component will make it possible to handle the explainability challenge by utilising the information, such as the path gathered from the knowledge graph, and visualising it.
✅ Adversarial Auditing of LLM-powered Search Engines
This project aims to build an evidence collection pipeline to scrutinise Large Language Model (L.L.M.) powered search engines transforming the internet search and discovery landscape.
It was already proposed for NGI. Search's 2nd call, but the jury's feedback mainly pointed to a need for more technical specifications. Since then, the pipeline has been prototyped, combining a prompt forgery, an experiment scheduler, browser automation and scraping modules, and an interface to explore and annotate the evidence.
✨This proposal aims to develop the tool for stable and feature completion, allowing open sourcing and enabling more researchers to conduct independent audits of BingChat, Google Bard, and ChatGPT-browsing mode.✨
The tool is already being leveraged to scrutinise the current elections in Germany and Switzerland. Initial results are compelling, with the first press articles expected to be published this October.
✅ SCION Browser
SCION is a path-aware inter-domain network architecture that provides applications and users opportunities to optimise data transport over the Internet. This project aims to integrate SCION into the Brave web browser to enable path-aware retrieval of web resources.
🔥However, finding the most suitable paths is a challenging problem🔥
This browser will use PANAPI to automatically find the corresponding paths, optimising application- and user-based metrics such as overall page load time, latency, bandwidth, privacy, and CO2 footprint according to the application's needs and user's preferences set in the browser. Additionally, it will also integrate support for RHINE into the Brave browser.
PS: RHINE is a name resolution protocol designed to provide an ideal naming service for the SCION Internet architecture. It enables better privacy and security properties for domain name lookup than existing DNS protocols.
✅ MetaVision
As a concept under emergence, the notion of the Metaverse varies. Yet there is a shared view on its benefits for digital art and making culture more accessible (e.g., transferring cultural information and content preservation).
The Metaverse provides a new metric for artists to be searchable, showcase/sell their artwork to a global audience with minimal barriers, and for consumers to have an unbounded experience not limited by physical space/entry fees.
Yet, a twofold problem remains: The actual creation of a (unique) 3D art gallery is a barrier for artists to showcase in a Metaverse setting, and trust (relating to digital content sharing) generates hesitation for posting artwork online.
Thus, this project will:
📌 Provide Free and Open-Source Software (FOSS) for artists to create Metaverse-ready art galleries procedurally.
📌 Investigate the integration of non-fungible tokens (NFTs) to ensure privacy and trust are upheld within the created emerging 3D web for art.
✅ TRUST4AI: Trustable AI-Driven Internet Search
Large Language Models (L.L.M.s) are emerging generative AI products that can produce novel and imaginative outputs. With the irruption of tools like ChatGPT—currently integrated into the search engine Bing— these models are expected to revolutionise how people search for information online and make decisions.
While the benefits of these systems are high in many aspects, they also pose key concerns regarding safety and trust. Initiatives to regulate the use of AI—most notably the European Union (EU.) AI Act—are expected to become crucial, forcing companies to commercialise AI products within the EU to ensure a certain level of trustworthiness.
However, checking an AI system's compliance with existing regulations is purely manual, making the process tedious, time-consuming, and unreliable.
✨This project aims to explore the potential of metamorphic testing in automating this task, making the process more reliable, affordable, and systematic.✨
✅ Space Cores
This project helps DevOps, DS, ML, AI, ETL, AV, and other engineering teams to find optimal instances for their batch jobs (e.g. "8 CPU cores, 64 GB of RAM, and a TPU needed in an EU datacenter to train ML models for 6 hours") by providing:
📌 Open-source tools, database schemas and documentation to monitor cloud and flexible VPS/dedicated server vendors and their compute resource offerings in an innovative and genuinely comparative way, including vendor details (e.g. location, certificates, green power), compute capabilities (e.g. CPU, memory, GPU/TPU), pricing (especially of spot instances), and performance (by running task-specific benchmarks).
📌 Managed infrastructure, databases, APIs, SDKs, and web applications to make these continuously and transparently tracked data sources publicly available and comparable in a validated, unbiased, structured, and searchable manner.
📌 Helpers can quickly start and manage instances at all the supported vendors with a standardised API.
✅ WASPER
This project seeks the support of the NGI. Search Programme to help our democracy by creating a tool for online publishers and users that will help us make informed decisions about the content we want to consume.
Their primary target is AI-generated texts, commonly and increasingly used by online trolls to spread misinformation, propaganda and manipulative claims.
They will focus on two aspects to develop this tool:
📌 Design of novel internally developed taxonomy that captures a more profound diversity of the types of trolling content;
📌 Apply state-of-the-art generative architectures with multilingual capabilities to be used in its development.
Their experienced data scientists will eventually produce tools to help media (and readers) detect trolling content created by AI - a primary target of challenge 2 of the Programme.
✅ EVA Gallery - The Virtual Space for Arts
The project's main objective is to develop a new platform for searching, finding, presenting, and supporting European visual art in virtual space for various audiences, from creators to gallerists and consumers.
Their vision proposes long-term sustainability and considerable potential for further developing and utilising cutting-edge technologies such as AI, virtual reality, and blockchain.
💡 Gallerists and artists can easily create their virtual galleries by creating and selling NFTs from their art. An AI assistant, E.V.A. (abbreviation for European Visual Arts), will help you identify and find the information and art pieces you seek.
✅ Indi
InDi aims to create an open-source search engine based on explainability, anonymity, fairness, and inclusiveness. InDi will use AI + NLP to re-rank results and give textual explanations. Yet, it will also rely on a community of voluntary human reviewers and validators to manually check/ accept/reject doubtful AI re-rankings, improving the InDi knowledge base for the following queries.
A blockchain mechanism will ensure a fair and anonymous review process, encouraging also to join reviewers from discriminated minorities. To sustain the best reviewers, InDi will also award them crypto-assets, utility tokens whose extrinsic value may ramp up as the platform takes momentum.
🤩 Participants, Cagliari University and R2M, have had a close-knit partnership for years and are leaders in their respective academic and business domains.🤩
✅ Better Food Search
Open datasets about food products and their ingredients (Open Food Facts, USDA FoodData Central, SmartLabel) and detailed information about ingredients themselves (Wikidata, Wikipedia) exist.
✨But while each provides ways to search its data, they propose to combine those datasets to create an LLM-powered search engine for food products. As part of the project, they will convert and combine datasets about food products and ingredients into semantic datasets.✨
For that, they will create a general tool to combine multiple datasets with the help of an LLM to clean the data (e.g., do entity resolution).
Once the unified dataset is made, they will use it to create an LLM-powered search engine, which will use it to support recommendations for food products based on factual data.
🫵 Ready to be the next NGI. Search trailblazer?
✨The N.G.I. Search 4th Open Call is still running, so you are still on time to join us in this journey to redefine the future of search and discovery!✨
👉 Learn EVERYTHING about this funding opportunity by clicking here
The NGI Search project is an endeavour for all who can convincingly demonstrate their commitment to transforming how we utilise, experience, explore, and uncover data and resources on the Internet and the web.
Remember to join the NGI Search Community and keep you updated by the project's Twitter (X) and LinkedIn for the latest news.
