<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Nadezda Chukhno on Medium]]></title>
        <description><![CDATA[Stories by Nadezda Chukhno on Medium]]></description>
        <link>https://medium.com/@nvchukhno?source=rss-33c97d1270b6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*365GZW_mS98SVUqvIpbMcQ.jpeg</url>
            <title>Stories by Nadezda Chukhno on Medium</title>
            <link>https://medium.com/@nvchukhno?source=rss-33c97d1270b6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 01:24:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nvchukhno/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[A First Look at How RAN Updates Disrupt Traffic Forecasting in Real Networks]]></title>
            <link>https://medium.com/@nvchukhno/a-first-look-at-how-ran-updates-disrupt-traffic-forecasting-in-real-networks-0d8d6fa1a3f7?source=rss-33c97d1270b6------2</link>
            <guid isPermaLink="false">https://medium.com/p/0d8d6fa1a3f7</guid>
            <category><![CDATA[radio-access-network]]></category>
            <category><![CDATA[measurement]]></category>
            <category><![CDATA[network]]></category>
            <category><![CDATA[traffic-forecasting]]></category>
            <category><![CDATA[5g]]></category>
            <dc:creator><![CDATA[Nadezda Chukhno]]></dc:creator>
            <pubDate>Thu, 26 Feb 2026 16:43:40 GMT</pubDate>
            <atom:updated>2026-02-26T16:47:06.212Z</atom:updated>
            <content:encoded><![CDATA[<blockquote><em>As researchers, engineers, and mobile network operators, we often talk about traffic dynamics, user behavior, and performance optimization — but we rarely talk about </em><strong><em>how often the</em></strong><em> </em><strong><em>network itself changes</em></strong><em>. </em><a href="https://dspace.networks.imdea.org/handle/20.500.12761/2006"><em>In this paper</em></a><em>, we seek to reply to a simple but largely unexplored question:</em> how frequently do operational Radio Access Networks (RANs) change, and what are the consequences of those changes?</blockquote><p><a href="https://github.com/nds-group/ran-updates">Using large-scale measurements</a> from a nationwide commercial network, we found that <strong>RAN updates are far more frequent and far more disruptive than commonly assumed</strong>, with important implications for data-driven network management and machine learning models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*joij4hAAZ9W3saLyf9EgTw.png" /><figcaption>This picture is generated by ChatGPT5.</figcaption></figure><h3>RANs are constantly evolving</h3><p>RANs are the backbone of mobile connectivity and the primary investment target for mobile network operators. To meet growing traffic demand, support new services, and roll out new radio technologies (e.g., 4G, 5G, 5G Non-Standalone (NSA), 5G Standalone (SA)), operators continuously upgrade their access networks.</p><p>Despite this, most existing studies implicitly assume that RAN deployments are static over the timescales used for traffic analysis and prediction. <em>Our measurements show that this assumption does not hold.</em> Analyzing multi-year configuration data for more than <strong>500,000 carriers </strong>(i.e., antennas), we observe that new carriers are deployed at a striking pace. At the national scale, <strong>new 4G carriers are opened roughly every 18 minutes</strong>, while <strong>5G carriers appear every 54 minutes on average</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zMdJ0ud4Eg0-6NuoAnQwjw.png" /><figcaption>Average time between open RAN carrier updates for 4G and 5G in France and across urbanization levels.</figcaption></figure><p>Even when zooming into cities, updates occur every few days — and in dense urban centers, multiple times per day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_comrhwh3NyaDp4qA9ILgw.png" /><figcaption>Average time between open RAN carrier updates for 4G and 5G within each of the ten major French cities.</figcaption></figure><p>In short, production RANs are highly dynamic systems.</p><h3>What happens during a RAN update?</h3><p>The vast majority of RAN updates we observe involve the <strong>deployment of new carriers</strong>, especially for 4G and 5G technologies. These carriers are usually added to existing sites rather than new locations, and they almost always reuse the same sector orientations already present at the site.</p><p>While this strategy is operationally sound, it directly affects traffic distribution. <em>When a new carrier becomes active, it starts serving part of the local demand, inevitably affecting the load on pre-existing carriers</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sNetxYnoXytDLcPo0AZLIg.png" /><figcaption>(a) Typical carrier life cycle observed in the RAN configuration data with transition times reported as median ± Median Absolute Deviation (MAD). (b) Types of RAN updates and their percent incidence on all modifications recorded for each RATs.</figcaption></figure><h3>Traffic is not just reduced — it is reshaped</h3><p>To quantify this effect, we analyze how carrier-level traffic changes following the activation of a new carrier. Focusing on clean update events, we find that <strong>over 7% (20,000+ carriers)</strong> <strong>of all 4G carriers nationwide experience statistically significant traffic changes</strong> due to these updates.</p><p>Among the affected carriers, three main behaviors emerge:</p><ul><li><strong>Traffic decreases</strong> (cluster A) for roughly half of the carriers, often sharply, as demand is offloaded to the new carrier.</li><li><strong>Traffic remains largely unchanged</strong> (cluster B) for about 40%, indicating that not all deployments have a strong local impact.</li><li><strong>Traffic increases</strong> (cluster C) for a smaller fraction, likely due to external factors such as demand growth or changes in nearby sites.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hRwI_dca1Kwry3X0GejOMg.png" /><figcaption>Centroids of the standardized daily time series of the traffic served by pre-existing carriers that are impacted by 4G (top) and 5G (bottom) open carriers updates, for each cluster.</figcaption></figure><p>Importantly, these updates affect not only traffic volume but also <strong>temporal patterns</strong>. Weekly traffic profiles before and after an update often differ substantially, meaning that the rhythm of network usage itself changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5n0ewOIP4iHKlneDw3RBUw.png" /><figcaption>Example of a RAN update event. The addition of two carriers operating in a new band at a site with two pre-existing 180◦ cells (top) impacts on (i) the real-world traffic at one pre existing carrier (middle) and (ii) the prediction accuracy of a N-BEATS model for the traffic at such a carrier (bottom).</figcaption></figure><h3>Why machine learning models struggle</h3><p>These findings have serious implications for machine learning–based network management. <strong><em>Many ML models are trained on historical traffic data under the assumption that future conditions will be similar to the past. </em></strong>RAN updates break this assumption.</p><p>We evaluate six widely used traffic forecasting models (e.g., DLinear, N-BEATS, LSTM, TCN, TSMixer, Transformer), train them on pre-update data, and measure performance before, during, and after a RAN update. Across all models and metrics, the outcome is consistent: <strong>prediction accuracy degrades significantly once the network changes</strong>.</p><p>On average, forecasting errors increase by <strong>41% after 4G updates</strong> and <strong>27% after 5G updates</strong>. In many cases, the degradation persists well after the update, <strong>indicating a lasting shift rather than a temporary anomaly</strong>. Thus, the state-of-the-art models struggle to adapt when the underlying infrastructure changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ROIxYhWmciYVOgt0k4rJkg.png" /><figcaption>Performance of ML models in terms of sMAPE and WAPE for control, pre-update, update, and post-update weeks.</figcaption></figure><h3>Rethinking “stationary” network data</h3><p>Our results highlight a fundamental mismatch between common research assumptions and operational reality. <strong>RANs are not static, and the traffic they carry is not stationary, even when user behavior remains similar</strong>.</p><p>This calls for a rethink of how we design and evaluate ML models for networks. Robustness to infrastructure changes, detection of distribution shifts, and rapid adaptation mechanisms are no longer optional features; they are essential for real-world deployment.</p><p>While proposing new models is beyond the scope of this work, we believe that exposing this phenomenon is a necessary first step. To support further research, we <a href="https://github.com/nds-group/ran-updates">release a labeled dataset </a>of carrier traffic traces affected by RAN updates.</p><h3>Closing thoughts</h3><p>RAN evolution is continuous, large-scale, and impactful. Ignoring it risks building models that perform well in controlled settings but fail in production. Our hope is that this work encourages the community to embrace the dynamic nature of real networks and to design machine learning models that can evolve along with them.</p><h3>Citation</h3><p>If you use <a href="https://github.com/nds-group/ran-updates">this dataset </a>(1,931 traffic time series measured at carriers affected by RAN modifications) and <a href="https://dspace.networks.imdea.org/handle/20.500.12761/2006">the paper</a> in your research, please cite:</p><pre>@inproceedings{ran_updates_2025,<br>  title     = {A First Look at Operational RAN Updates and Their Impact on Carrier Traffic Demands and Prediction},<br>  author    = {[Antonio Boiano, Nadezhda Chukhno, Zbigniew Smoreda, Alessandro E. C. Redondi, Marco Fiore]},<br>  booktitle = {[INFOCOM 2026]},<br>  year      = {2026}<br>}</pre><h3>Acknowledgement</h3><p>Nadezda Chukhno’s work is funded by the European Union, Grant Agreement number 101206327 (6G AI-TANGO). The work was also funded by CoCo5G (ANR-22-CE25–0016) of the French National Research Agency (ANR) as well as by the 6G-IRONWARE (CNS2023–143870) funded by MICIU/AEI /10.13039/501100011033 and by the European Union NextGenerationEU/PRTR, and by the 6G-AI-TANGO (GA 101206327) funded by the European Union.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zucEWveN1TI0iWL2hac0Xw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bX1wuOM0NiZoTh9DRWgt-A.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0d8d6fa1a3f7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MSCA PF: tips on how to write a successful proposal (personal story)]]></title>
            <link>https://medium.com/@nvchukhno/msca-pf-tips-on-how-to-write-a-successful-proposal-personal-story-72cbe0be77fa?source=rss-33c97d1270b6------2</link>
            <guid isPermaLink="false">https://medium.com/p/72cbe0be77fa</guid>
            <category><![CDATA[proposal-writing]]></category>
            <category><![CDATA[postdoctoral-researchers]]></category>
            <category><![CDATA[funding-application]]></category>
            <category><![CDATA[msca-fellowship]]></category>
            <dc:creator><![CDATA[Nadezda Chukhno]]></dc:creator>
            <pubDate>Mon, 08 Sep 2025 13:05:19 GMT</pubDate>
            <atom:updated>2025-09-08T13:05:19.531Z</atom:updated>
            <content:encoded><![CDATA[<p>In 2024, I wrote a proposal for the MSCA PF (European) and received it the same year. My first attempt was a success. I gathered a lot of feedback while writing the proposal. The process was long, and I learned a lot, and now I’d like to share some writing tips that may also be helpful to you.</p><p>My proposal is about practical AI for Time-vAryiNG network traffic fOrecasting in 6G (6G-AI-TANGO) → So my first suggestion is to try to create a nice acronym for the project that would ‘stick’ to the mind of the evaluator. Later, I even designed a logo for the project; you can see it at the end of the blog.</p><p>Here is my proposal´s evaluation score:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g3Dhpf3cZLI02j09hyhFTw.png" /><figcaption>My evaluation scores.</figcaption></figure><p>My main suggestion to ensure your proposal succeeds is to <strong>address every element outlined in the template</strong>. Each question is accompanied by a list of specific items they expect to see, which are the components that ultimately earn points. The template is highly detailed, and for each question, it clearly specifies what you need to write.</p><p>Write clearly and use simple words to <strong>explain how the training will benefit you</strong>. The template will show you what to answer. How you express this is personal and depends on the topic and your background. For example, I mentioned that I have no experience in designing new ML models, which they could see from my publications. I also mentioned some methods I plan to learn and apply, and explained who will train me. It’s important to write in simple language because the reviewer might not be an expert in your field. Your writing should convince both non-experts and experts.</p><p>One significant challenge I faced was<strong> condensing everything into 10 pages</strong>. Based on feedback from the 2 individuals with whom I shared the proposal, I received suggestions on what could be removed or improved. Although I only shared it with two external people (besides my supervisor), I’ve heard that some people share it with many more to gather additional opinions. However, based on my experience, more than 2 opinions might be too much to handle, but it also depends on the quantity of comments you would receive. Regardless, I strongly recommend <strong>sharing a complete draft with at least one person outside the project and team</strong> — those who have no prior knowledge of the proposal and are possibly not very familiar with the topic, however, with experience writing MSCA proposals. They will provide the most valuable suggestions.</p><p><strong>Write actively</strong> so the reviewer is not bored, and try to create a <strong>visually appealing structure</strong>. 10 pages are too tight, but it is not nice to read just the text. Separate it into smaller paragraphs, use bullet points, pictures, tables, bold and/or italic fonts to highlight important stuff.</p><h3><strong>Layout tips</strong></h3><ol><li>Avoid having paragraphs that are too long, as they are hard to read. <strong>Prefer shorter paragraphs</strong> that contain one main concept.</li><li><strong>Use lists</strong> whenever this makes the reading more fluent/interesting. You may lose a bit of space there, but the document is more readable to the evaluator.</li><li>Do not highlight in bold individual words, but <strong>whole sentences </strong>that are self-contained. The evaluator may <strong>only read</strong> the bold text, and he/she should still grasp the high-level idea of the paragraph/section!</li><li>Format figures better to save space (wrap them around the text), and add captions, etc.</li></ol><h3><strong>Main resources that can help you</strong></h3><p>I believe it’s important to start with the resources that helped me a lot when writing the proposal, as well as understanding where to start.</p><ol><li>Download a <a href="https://circabc.europa.eu/ui/group/9a4230a6-2769-4067-b136-3e238ae65fa6/library/96b41c46-5aeb-4c96-89fc-df9d76a2da2b/details">guide for applicants</a>.</li><li>The proposal template can be found once submission has been started and a proposal profile is created on the <a href="https://ec.europa.eu/info/funding-tenders/opportunities/portal/screen/home">Funding &amp; Tender Opportunities Portal</a>. When you start the application, ensure that you select the correct type of MSCA PF. <em>e.g.</em>, European Fellowships (EFs) or Global Fellowships (GFs).</li><li>Search for the <a href="https://euraxess.ec.europa.eu/worldwide/china/news/handbook-msca-pf-call-2025-released">MSCA POSTDOCTORAL FELLOWSHIP HANDBOOK</a> (preferably the same year as the application, as the requirements might vary from year to year). It has been prepared as a transnational cooperation between National Contact Points (NCPs) for the MSCA, so a great deal of MSCA expertise is behind it.</li><li>Search for tips from MSCA PF fellows, as some people share their experience (like me now ☺).</li></ol><p>In this blog, I won’t repeat what’s in the guidelines and handbook. I’ll share what worked for me.</p><h3><strong>Parts of the MSCA PF proposal</strong></h3><p>Be ready to prepare the following 3 parts:</p><ul><li><strong>Part A </strong>is mostly about administration. It includes forms that need to be completed online.</li><li><strong>Part B1 (max 10 pages) is the part evaluated by reviewers. </strong>In this blog, we discuss this part.</li><li><strong>Part B2 </strong>complements B1 and helps in the evaluation. It includes the applicant’s CV, details about the participating organization(s) and their capacity, and the commitment of any associated partner(s), if relevant.</li></ul><h3><strong>Main sections of the research proposal (B1 part)</strong></h3><p>There are <strong>three main sections</strong> of the B1 part, which are</p><ol><li>Excellence (weight 50%)</li><li>Impact (weight 30%)</li><li>Implementation (weight 20%)</li></ol><p>Part B1 of the proposal can only be 10 pages long. Most agencies suggest using a 5–2–3 structure. My first draft was almost 7–1–2, which was not balanced. I finally got it to 5.8–1.7–2.5. It’s important to cover everything asked (which is a lot), and it might seem repetitive between sections. But this repetition helps link all three parts of the proposal. Let’s look at each section in detail.</p><h3><strong>1. Excellence</strong></h3><p>In this part, you need to describe how strong and innovative the research idea is, the quality of the training and knowledge transfer, and the researcher’s potential. For me, this was the most difficult part. It is divided into 4 subsections:</p><p><strong><em>1.1. Quality and pertinence of the project’s research and innovation objectives (and the extent to which they are ambitious, and go beyond the state of the art).</em></strong></p><p>Include sections like 1.1.1. Introduction (or Context and motivation), 1.1.2. State-of-the-art (SOA) and research gap, and 1.1.3. Research objectives (ROs). Adding figures can help readers understand better. In context and motivation, clearly explain the main idea of the project.</p><p>It is recommended in the introduction (but I did it after defining ROs) to explain <strong>how the project aligns with the H2030 agenda, sustainable goals, or other European Commission initiatives</strong>. It’s important to show that the topic is of interest to the European Commission and explain who will benefit from the project results. This shows the project’s impact. Because <strong>no impact means no money</strong>. For example, I included this line:</p><blockquote>Moreover, making AI models robust to temporal variations would contribute to the AI Strategy proposed by the European Parliament, thereby facilitating further AI development.</blockquote><p>If your supervisor or you did the SOA work, do state so. Give a clear summary of what was done and point out the research gaps. Use footnotes for references to save space. The footnote size can be 8 points, while the main text font is 11 points.</p><p>In section 1.1.3. Research objectives (ROs), state research objectives that will fill the identified gaps. Remember, ROs should be SMART:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hg3Ad5wZdot5D08YxSZ9dQ.png" /><figcaption>SMART objectives.</figcaption></figure><p>To fully address SMART criteria and in addition to examples I described in figure above, I have also added the following:</p><blockquote>Planned duration for RO1 activity is 18 months, while RO2 and RO3 are up to 12 months each, as detailed in Section 3.1. Objectives may overlap within timeline to be adapted quickly to changes. Concept drift, a known term in ML for different prediction tasks, e.g., image and text classification, ensures the feasibility of ROs for network traffic KPI forecasting with some necessary adaptations.</blockquote><p>Also, make sure to <strong>clearly emphasize what is innovative and original</strong>. If your work builds on previous research from your group, highlight the specific new angles, contributions, or differentiators that set your work apart!!!</p><p>It is a good idea to state the project’s goal before listing the research objectives. For example, I did it this way:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LM56kV9mOLnBuLnyc1n5fA.png" /></figure><p><em>Checklist for this subsection:</em></p><p>— The quality and pertinence of the research and innovation objectives are well-defined, well-aligned with current tech trends (in my case, virtualisation and AI-driven management).</p><p>— The research objectives outline innovative, with <strong>measurable performance indicators</strong> for tracking progress.</p><p><strong><em>1.2. Soundness of the proposed methodology (including interdisciplinary approaches, consideration of the gender dimension and other diversity aspects if relevant for the research project, and the quality and appropriateness of open science practices).</em></strong></p><p>The methodology explains how you will reach your goals. It focuses on the scientific ideas and resources. You need to highlight what makes your research approach special.</p><p>I split the methodology into 3 stages (each one is associated with a WP and RO). This stage-RO-WP connection should be clearly stated, <em>e.g.,</em></p><blockquote><strong>Stage 1: Assessing and quantifying</strong> temporal drifts in network KPI forecasting (RO1, WP1).</blockquote><blockquote><strong>Stage 2: Designing practical solutions </strong>for time-resilient network AI predictors (RO2, WP2).</blockquote><blockquote><strong>Stage 3:</strong> <strong>Evaluating and validating </strong>time-resilient AI models in realistic scenarios (RO3, WP3).</blockquote><p>For each stage, I specified the tasks and exact methods that I will apply for each task. It is important to note that my stages are mapped 1:1 to the research objectives (even names are very similar). In the guidelines, they ask to describe and explain the overall methodology, including concepts, models, and <em>assumptions</em>.</p><p><em>Use italic to make the task titles more evident.</em></p><p>A reviewer might not be an expert in your field. Try to use simple explanations and provide details on why and what you need the mentioned methods for. In my case, <em>e.g.,</em> I explained how adaptive learning is different from online learning, and why I need both in my analysis. Also, describe your assumptions and <strong>talk about any big challenges in the methods you might face and how you plan to solve them</strong>.</p><p>If you use AI, you must demonstrate its <em>technical robustness, </em>which includes many things — see page 18 of the MSCA-NET handbook.</p><p>Furthermore, <strong>highlight the interdisciplinary nature of your research, including its gender dimension, open science practices, and FAIR principles</strong>, among others. For example, in terms of interdisciplinarity, the research conducted within my project is by its own nature highly interdisciplinary, primarily building on domain knowledge in mobile networking but combining that with competencies in ML, particularly advanced deep learning models. It requires proficiency in AI, wireless communications, measurements, data analysis, and mathematics to understand and implement different ML algorithms.</p><p>If gender and diversity aspects are irrelevant and the research findings are purely technological, do state so.</p><p>It is very important to make research results openly accessible to the public using the FAIR principles (Findable, Accessible, Interoperable, Reusable). You will explain how to do this later in your data management plan. Here is an example of a line from my proposal:</p><blockquote>The research results will be made available via publishing preprints and peer-reviewed top-tier IEEE/ACM journals with open access, as well as on my personal webpage (link available in CV), ResearchGate, Zenodo; the code will be available on my GitHub page to ensure reproducibility and reusability via FAIR principles. More details will be provided in the data management plan (DMP). At least one publicly available dataset will be used in addition to private ones (due to confidentiality agreement).</blockquote><p><em>Checklist for this subsection:</em></p><p>— The project employs a robust interdisciplinary methodology that addresses key challenges with advanced techniques (and real-world data, in my case, was an advantage), promising impactful results.</p><p>— The proposal clearly articulates how approaches from different disciplines will be properly combined (in my case, by integrating AI in 6G networks).</p><p>— The methodology commits to open science by ensuring, for example, source code availability and transparent result sharing on platforms like ResearchGate and GitHub, fostering collaboration and enhancing project impact through effective knowledge dissemination.</p><p>— The proposal includes technically robust AI systems, ensured by a structured methodology that emphasizes data quality, adaptability, and comprehensive evaluation.</p><p><strong><em>1.3. Quality of the supervision, training and of the two-way transfer of knowledge between the researcher and the host.</em></strong></p><p>In this subsection, the supervisor’s experience in supervising and their relevant expertise in the project field should be clarified. His/her expertise is important for the evaluation of the proposal.</p><p>Further, the important thing is to highlight what <strong>your added value is to the institution </strong>and <strong>what added value your host institution provides </strong>to you and your career <em>(</em><strong><em>two-way transfer of knowledge)</em></strong>. Concentrate also on your training opportunities (both research-related and management-related) as MSCA PF is a training program.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gNC2xetsijB7sTDRVz3e7g.png" /><figcaption>Two-way transfer of knowledge. The picture is generated by ChatGPT5.</figcaption></figure><p>It is always good when the researcher and the host complement each other.</p><p>In case of non-academic placement, the rationale for non-academic placement should be provided in this section.</p><p>Here are more details about the secondments and non-academic placements:</p><ul><li>The <strong>secondment(s) duration is up to one-third </strong>of the standard project duration (2 years for European MSCA) and is conducted <strong>during</strong> the project.</li><li>The <strong>non-academic placement duration is up to 6 months,</strong> an <strong>additional period</strong> that occurs only after the standard duration of the fellowship.</li></ul><p>Interdisciplinarity is highly valued. From my experience, collaboration between academia and industry (as in the case of non-academic placements that also extend your fellowship by 6 months) is highly valuable, as it demonstrates that the research you are conducting is practical.</p><p><em>Checklist for this subsection:</em></p><p>— <strong>The supervisor’s expertise in the field and strong mentorship record align with the project’s goals</strong>, ensuring effective support,<em> e.g.,</em> in my case, through industry connections and unique datasets.</p><p>— The<strong> bidirectional knowledge transfer </strong>between the researcher and host institutions, including non-academic placement, equips the researcher with useful stuff, <em>e.g.,</em> with skills in resilient AI predictor design and access to unique mobile network traffic datasets, advancing telecommunications and AI.</p><p>— The non-academic placement enhances value by enabling real-world validation of project solutions.</p><p>— Training activities ensure effective knowledge transfer, merging scientific rigor with practical applications and fostering interdisciplinary collaboration (<em>e.g.,</em> in telecommunications and AI).</p><p><strong><em>1.4. Quality and appropriateness of the researcher’s professional experience, competences and skills.</em></strong></p><p>This subsection is quite simple. I summarized my research experience, a strong background in wireless communication research, with expertise in algorithm design, optimization, and applied ML. My PhD work, carried out under the EU Horizon 2020 Marie Curie programme <em>(mention if you were a part of MSCA before!)</em>, tackled complex problems, leading to publications in high-impact venues. Beyond my doctoral research, I’ve gained experience in collaborating with both academia and industry, and contributed to applied innovations such as invention disclosures. In addition, I’ve contributed to the training and supervision of students, published extensively, and built a track record of interdisciplinary and international collaboration that bridges theoretical research and practical applications.</p><p>And finished with the summary, concluding <strong>why I am the best for this project</strong>:</p><blockquote>My experience in mobile networking, AI, data analysis, mathematics, and simulations resulted in 750+ citations and an H-index of 13 (source: Google Scholar). This expertise is essential for mobile network analysis, extracting insights from network data, and leveraging and adjusting ML models to design robust solutions, positioning me ideally to tackle the project’s challenges.</blockquote><p><em>Checklist for this subsection:</em></p><p>— Though still in the early stages of their career, the researcher demonstrates a strong academic foundation and theoretical knowledge in the field, which positions them well for this type of research.</p><h3>Weaknesses of my proposal in Section 1, or what can go wrong</h3><p>The weaknesses (or mistakes) are something we learn from. So, I think it is useful to state what was wrong with my proposal (I got 4.8 out of 5) in this section.</p><p>— <em>The approach to solve the generalizability of existing AI-driven predictors to different conditions is not well addressed.</em> This relates to the research methodology/approach (1.2. soundness of the research methodology). It signals that evaluators felt the proposal did not clearly explain how my methods would ensure robustness and transferability of AI models across different scenarios.</p><p>— <em>The researcher’s 6G experiences are not sufficiently highlighted. </em>This refers to the researcher’s expertise and capacity to carry out the project (1.4. quality and appropriateness of the researcher’s experience). Evaluators expected to see stronger emphasis on my past work and achievements in 6G, which strengthens credibility and fit.</p><p>Be sure to address those moments properly in your proposal.</p><h3><strong>2. Impact</strong></h3><p>In this part, they ask how the fellowship will boost the researcher’s career, strengthen collaborations, and create broader benefits for science, economy, and society.</p><p>In this section, I lost most of the points. Remember, <strong>no clear impact, no money!</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f240fysTX7w5oSK5fRNsvQ.png" /><figcaption>Impact illustration. This picture is generated by ChatGPT5.</figcaption></figure><p>The section is divided into 3 parts:</p><p><strong><em>2.1 Credibility of the measures to enhance the career perspectives and employability of researchers and contribution to their skills development.</em></strong></p><p>You should explain <strong>career goals after the fellowship</strong> (academic or industry), how the project helps you reach them, and what <strong>skills and experiences</strong> you’ll gain that future employers will value. Highlight the <strong>impact and sustainability of collaborations</strong> (academic, industrial, interdisciplinary), and if relevant, the role of <strong>secondments or placements</strong> in strengthening your career prospects and employability.</p><p><em>Checklist for this subsection:</em></p><p>— The outlined measures are well-suited to enhancing the researcher’s career perspectives and employability, making them a strong candidate for future opportunities in both academia and industry.</p><p>— The proposal offers valuable skills development opportunities for the researcher, in my case, including collaboration with mobile network operators through non-academic placements. This approach bridges academia and industry, fostering essential connections and practical experience that will benefit the researcher’s future career in both sectors.</p><p><strong><em>2.2. Suitability and quality of the measures to maximise expected outcomes and impacts, as set out in the dissemination and exploitation plan, including communication activities.</em></strong></p><p>I articulated this section 2.2 with three main blocks.</p><p><em>2.2.1 Ambitious dissemination of the project results to target (expert) audiences and open-access materials</em></p><p>I enumerated the journal and conference papers expected in the action and provided an itemized list, 1 entry per paper, indicating the main contribution and the expected venue. In fact, the GANTT diagram in the next section should support it. That’s it, simple,<em> e.g.,</em></p><blockquote>I expect submission of minimum four journal publications in top ISI-indexed journals covering multiple disciplines, which will be published as open-access by taking up self-archiving rights or paying the fees. The early versions of papers will be published on ArXiV. Foreseen journal papers (JP) with corresponding contributions are:</blockquote><blockquote>• JP1 — Temporal drifts analysis — IEEE Networking Letters (new journal)</blockquote><blockquote>• JP2 — Resilient AI design based on re-training — IEEE Transactions on Mobile Computing (IF 7.7)….</blockquote><p>Instead of listing the conference papers (you can do it), I told that</p><blockquote>Each journal will be preceded by a presentation at a top-tier conference, such as IEEE INFOCOM, ACM CoNEXT, ACM MobiCom, and AAAI, respectively, to gain feedback and enhance visibility.</blockquote><p>Then, you can provide a table with all the seminars you will organize. Activity, Targetted audience, Month within your MSCA, Location, Metrics (expected audience). I included only seminars, but you can also specify special sessions in conferences, workshops within conferences, and tutorials, etc. You can mention publishing datasets and other ways to disseminate.</p><p><em>2.2.2 Exploitation of the results, Intellectual Property, and standardization efforts</em></p><p>Mention the contacts to the industry that you and the host institution can bring, both!. Give the context about open science and patenting.</p><p>Here is an example:</p><blockquote>The project adheres to the premise “As open as possible, as closed as necessary.” The ownership on Intellectual Property Rights (IPR) will be retained at HOST institution. The IPR issues, including foreground and background protection, will be signed individually with the non-academic placement and short visit unit, and other companies interested in the commercial exploitation of 6G-AI TANGO developments. All agreements will be developed in line with the recommendations of the EC, following the rules and regulations of the EC Grant Agreement. Thus, the commercial exploitation of the results will take place in conformity with the interests of the EU, in accordance with the aims of the H2020 programme objectives, and with the Open Science and Open Source policy of the action. Commercial exploitation is the logical project follow-up, which will enable an alternative funding source after the MSCA-PF action. We have already identified two technological advances which could be subject to IPR protection:<br>Patent 1: Provide a name for the patent<br>Patent 2: Provide a name for the patent</blockquote><p><em>2.2.3 Quality of the proposed measures to communicate the project activities to different target audiences</em></p><p>Here is where you have to provide a list describing each way to communicate the project,<em> e.g.,</em> Website, Social Media, YouTube, Flyers, Local Mass Media, and European Researchers Night. You can also mention, <em>e.g.,</em> how many subscribers you plan to reach.</p><p><em>Checklist for this subsection:</em></p><p>— The suitability and quality of the measures to maximize expected outcomes and impacts, as outlined in the dissemination and exploitation plan, are excellent, and the target groups are well addressed.</p><p>— The planned communication and public engagement activities are highly suitable, with well-defined objectives, clear messaging, and an effective selection of tools and channels.</p><p>— The strategy for the management and protection of IP is highly suitable, ensuring ownership is retained while collaborating with various companies.</p><p><strong><em>2.3. The magnitude and importance of the project’s contribution to the expected scientific, societal and economic impacts.</em></strong></p><p>In this subsection, you need to provide concrete scientific, societal, and economic impacts. It is better if you can provide specific values. In my case, scientific impact was quite straightforward, but in general, this section was not convincing enough. Therefore, there is no c<em>hecklist for this subsection. </em>Instead, we will have a look at my mistakes.</p><h3>Weaknesses of my proposal in Section 2, or what can go wrong</h3><p>— The societal impact is presented in a generic way and lacks sufficient elaboration.</p><p>— The magnitude and importance of the expected impacts are insufficiently convincing.</p><p>In short, you should <strong>move from generic statements to concrete, evidence-based impacts</strong> that are explicitly connected to European/global challenges and to your career development.</p><h3><strong>3. Implementation</strong></h3><p>This part is about how well the project is planned and managed (work plan, resources, supervision, institutional support, feasibility).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qynVvFRvQPVVnPOCJVsSJQ.png" /><figcaption>Implementation section illustration. The picture is generated by ChatGPT5.</figcaption></figure><p>This section is divided into 2 parts:</p><p><strong><em>3.1 Quality and effectiveness of the work plan, assessment of risks and appropriateness of the effort assigned to work packages.</em></strong></p><p>Be sure that</p><p>— Work packages are well- and logically-organized. Proposed tasks are meticulously aligned with the objectives. Clear milestones and deliverables significantly enhance accountability and tracking;</p><p>— The effort distribution across work packages demonstrates outstanding precision and careful planning. The effort assigned to each work package is both appropriate and highly effective in supporting the project’s objectives.</p><p>Here are the examples from my proposal:</p><blockquote><strong>WP 2. Designing practical solutions for time-resilient network AI predictors</strong>. All WP2 tasks will proceed in parallel, starting 9 months after the project kick-off meeting and ending at M24. The rationale is that the work carried out in these tasks is not disjoint, but likely, the results in one of the tasks will affect progress in others (e.g., identifying a new effective few-shot learning approach in T2.2 may reduce the cost of re-training and thus impact on T2.1).</blockquote><blockquote><strong>Result T2.1: </strong>Network AI predictors capable of continuous self-adjustment to time variations.</blockquote><blockquote><strong>Result T2.2: </strong>Network AI predictors with few-shot learning capabilities in new contexts.</blockquote><blockquote><strong>Result T2.2: </strong>Network AI predictors based on time-resilient GNN architectures.</blockquote><blockquote><strong>Milestone: Mi3</strong> [M18] is reached when network AI predictors are ready for evaluation.</blockquote><blockquote><strong>Deliverable: D2</strong> — summary of the designed time-resilient AI models [M18].</blockquote><p>In addition to the first 3 technical WPs, I also included WP4: Management, WP5: Training and transfer of knowledge, and WP6: Dissemination, exploitation, and communication. There is a list of compulsory deliverables, <em>e.g., </em>Data Management Plan, Carrier Development Plan, and Plan for Dissemination and Exploitation. They should be mentioned; check the handbook for more details.</p><p>— The Gantt chart is excellent, because it contains all elements of a well-planned research work. It reflects exceptional foresight and a robust understanding of project timelines and dependencies, <em>e.g.,</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ov-aBkqlYEpdLyoOw3q6Mw.png" /><figcaption>Example of the Gantt chart.</figcaption></figure><p>Also, it might be a good idea to illustrate the interconnection among WPs, <em>e.g.,</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1008/1*0sFKhaDDJdbekTAOlu-URg.png" /><figcaption>Interconnection among WPs.</figcaption></figure><p>Be also sure that</p><p>— The proposal provides a comprehensive and well-articulated risk management strategy. The key risks are thoroughly considered, and effective mitigation measures are outlined in a clear and detailed table format, ensuring that potential challenges are managed effectively to safeguard project objectives.</p><p>For the risk Table, I included Risk description, Impact (low, medium, high), Probability (low, medium, high), and Mitigation strategy. Try to use a narrative description.</p><p><strong><em>3.2. Quality and capacity of the host institutions and participating organisations, including hosting arrangements.</em></strong></p><p>In this subsection, you have to ensure that</p><p>— The proposal details <strong>exceptionally strong hosting and research arrangements supported by a high-capacity institution</strong>. The general hosting, research arrangements, and institutional support are presented in detail, demonstrating a high level of commitment and promise for the successful execution of the project.</p><p>— The quality and capacity of the host institution and the non-academic placement institution are outstanding and fully consistent for the realization of the proposed activities, <strong>ensuring an ideal environment for the realization of the project’s goals</strong>.</p><h3><strong>Final Tips</strong></h3><ol><li>Send your proposal to experienced individuals; they will provide you with their perspective.</li><li>Check all the criteria listed in the template and handbook. Check their <strong>point-by-point directions</strong> on each heading and subheading, and also notice the most common mistakes in each section (strengths vs weaknesses). This document (handbook) provides extensive information and additional support for you.</li><li>Interdisciplinarity is highly valued.</li><li>Be coherent, use good language, and have a pretty layout.</li><li>Enthusiasm is important: Write actively!</li><li>Having a successful proposal from previous years is very helpful.</li><li>Write, rewrite, get feedback, modify, improve, check criteria, rewrite.</li></ol><p>Good luck!🤞Remember, your proposal is not just about ticking boxes — it’s a chance to share your vision with the world and bring some impact. Believe in your ideas and let them shine!</p><h3>Special thanks to</h3><p>Ximo and Livia for their valuable comments, to Marco for his guidance, and to Olga and Tom for their constant support and help.</p><h3>Acknowledgement</h3><p>Nadezda Chukhno’s work is funded by the European Union, Grant Agreement number 101206327 (6G AI-TANGO).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zucEWveN1TI0iWL2hac0Xw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bX1wuOM0NiZoTh9DRWgt-A.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=72cbe0be77fa" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>