Niki Athanasiadou, PhD
New York, New York, United States
4K followers
500+ connections
About
I have practical experience and a thorough understanding of big data analytics, from…
Activity
4K followers
-
Niki Athanasiadou, PhD posted thisSaturday ramble 🚶♀️ 🌳 ☀️ ➡️ From ecology: the modern human does not have any natural predators. ➡️ Logical consequence (in theory): the probability of dying a violent death should be zero. ➡️ Facts: They are not (https://lnkd.in/eqTYevvj) 👆 In this random morning "mental experiment", I was able to find public data and satisfy my curiosity. But the best data in my opinion (linked above) were in textual format, not easily parsable for plotting and further analyses. This is only one article, but what if we had thousants or millions documents we wanted to extract data from? ⭐ Data extraction, and turning unstructured information into structured data, might be the most valuable, unambiguous application of LLMs. Let me know what you think! #data #notRAG #LLM #unstructured #structured
-
Niki Athanasiadou, PhD reposted thisNiki Athanasiadou, PhD reposted thisΤο Τμήμα Βιολογίας είναι στην ευχάριστη θέση να ανακοινώσει την επίσημη σελίδα του στο LinkedIn. Προσκαλούμε όλους τους φοιτητές, τους αποφοίτους, τους συναδέλφους και τους συνεργάτες μας να συνδεθούν! School of Biology is happy to announce its official page in LinkedIn. We invite all of our students, alumni, colleagues, and collaborators to connect!
-
Niki Athanasiadou, PhD shared this📣 If you are looking for a Postoc and want to work in a great team on an exciting project, I urge you to apply! #bioinformatics #computationalbiology #postdocNiki Athanasiadou, PhD shared thisStarting 2026 with an exciting new job? Please note that the deadline has been extended for the opening for a new postdoc position in our team at the Victor Phillip Dahdaleh Heart & Lung Research Institute, University of Cambridge Department of Medicine Get in touch to join the Cambridge Cardiovascular community https://lnkd.in/eigYD7Ee
-
-
Niki Athanasiadou, PhD shared thisTruth can be funny too! For all #scientists <3Niki Athanasiadou, PhD shared this"They’re often rather unreasonable people, you know that? Reasonable people don’t discover the North Pole. They don’t discover the South Pole. They don’t cross deserts, climb mountains, and they don’t discover, to a first approximation, things in science. Most of us scientists are in some way unreasonable people who don’t fit the mould." - Scientist and medicine laureate Peter Ratcliffe's best advice to young researchers Ratcliffe was awarded the 2019 medicine prize for "discoveries of how cells sense and adapt to oxygen availability.”
-
Niki Athanasiadou, PhD shared thisWe often forget that the modern scientific method is a product of philosophy, not engineering. Thanks for reminding us Justin Bélair 👏 👏👏👏Niki Athanasiadou, PhD shared thisWhy we NEVER prove the null hypothesis (H0)? Because there is an asymmetrical logical relationship between knowledge and data : it takes an infinite amount of data to confirm that a hypothesis is “ALWAYS true”, while it only takes one contradiction to affirm that a hypothesis is “NOT ALWAYS true”. In philosophy of knowledge, this is called the induction problem and is often illustrated with the Black Swan Example : - No matter how many white swans have been observed in the past, we can never be 100% certain that “All swans are white” is TRUE. - But if we flip the problem the other way, we can observe a SINGLE black swan and logically conclude with perfect rigour that “All swans are white” is FALSE. This falsification approach to scientific knowledge was put forth in philosophy by Karl Popper in 1934. Yet, statisticians such as Karl Pearson and Ronald A. Fisher were early to the party in developing Significance Testing and Hypothesis Testing from a statistical point-of-view decades earlier – it was in l’air du temps (the spirit of the times, the Zeitgest), as we say in French. Have you made any interesting connections from philosophy of science and statistics? Please share below! In another life, I was a philosophy enthusiast 😇
-
Niki Athanasiadou, PhD shared thisPlot twist: it’s been the data all along, genious!Niki Athanasiadou, PhD shared this“If you booted up a super-smart AI in ancient Greece, fed it all human knowledge, and asked it how to land on the moon, it would respond “You can’t land on the moon. The moon is a god floating in the sky.” How would you get it to realize the moon is actually a big rock? That’s a great, poorly defined problem, and I don’t expect AI to solve it anytime soon.” https://lnkd.in/eb5YjZAn
-
Niki Athanasiadou, PhD shared thisLife is too short for only serious posts. Happy Sunday to the #datascience #machinelearning #AI community! 😄Niki Athanasiadou, PhD shared this"𝗙𝗮𝘀𝗰𝗶𝗻𝗮𝘁𝗶𝗻𝗴!" "𝗕𝗿𝗶𝗹𝗹𝗶𝗮𝗻𝘁𝗹𝘆 𝗮𝗿𝘁𝗶𝗰𝘂𝗹𝗮𝘁𝗲𝗱" "𝗗𝗲𝗲𝗽𝗹𝘆 𝗶𝗻𝘀𝗶𝗴𝗵𝘁𝗳𝘂𝗹" This paper investigates the perceived emergent intelligence in Large Language Models (LLMs) and derivative generative AI systems. The author claims that the popular characterization of AI output as “magical,” “brilliant,” or “sentient” is not a function of the system’s capabilities, but rather a property that emerges precisely at the point where the observer’s explanatory competence collapses. In simple terms: AI’s magic starts at the edge of our own stupidity. For the purpose of this paper, we will refer to this theoretical threshold as the ‘Edge of Stupidity’ (EoS). And it’s precisely at this boundary that critical thinking ceases to function and is replaced with naive awe. Link to pre-print paper: https://lnkd.in/eaMCzcgB (And yes, if it you didn't pick up on it already, this is satire. Don't forget to smile today 🫶) #notarealpaper #haveagoodday #ai
-
Niki Athanasiadou, PhD shared thisTruth be told...generative LLMs are math not magic 😉 Thanks Dagmar Monett for pointing to this gem of a post!Niki Athanasiadou, PhD shared thisSome "AI" vocabulary: - LLM: Large Language Model. A statistical pattern machine that does NOT "know" things. They're just next-token predictors trained on existing texts. Ask a question and they do NOT check a database. They spin a statistical wheel of fortune. - Prompt: Your input. For example: "Marie Curie is remembered for…" Most of the time you'll get "radioactivity" or "Nobel Prize". You might also get "discovered penicillin". Wrong. But it sounds right, because in the vector space of the LLM, "scientist" lights up near "medical breakthrough" and "discovery" exist near "penicillin". The model MAY stitch it together into a fake fact. Confidently delivered, however. - Hallucination: from the model's perspective, this so-called "hallucination" is IDENTICAL to the truth. Same patterns. Same confidence. There is no model of confidence level, no warning light to the user. Because the model cannot know the distinction. It just predicts the next token. - RLHF: Reinforcement Learning from Human Feedback. This is what we sometimes call lipstick on the pig. Humans do NOT teach the model "truth" (because, remember, it cannot model truth), we're teaching polish. The model learns to hedge: "While I cannot verify…" It looks cautious and aware, but it's still guessing. Just smoother guessing. Will it be good enough for your particular service? Perhaps. But you (and your poor client that ordered "AI" from you) need to be aware of the limitations. From a truth perspective, RLHF might actually worsen the user experience because of the hand-wringing hedging. - RAG: Retreival-Augmented Generation. This is the industry's favorite patch: bolt on a traditional search engine and some internal documents; give the model a cheat sheet. It helps, a little. But the generator is still the same. It can misquote, remix, or confidently hallucinate on top of the retrieved facts. Garbage in, polished garbage out. And while researchers have warned us about this for YEARS, it has finally dawned on the biggest vendors (like OpenAI) that hallucinations aren't going away. Ever. They’re a feature of the architecture, not a glitch (see link in comments). LLMs are pattern machines. They're great at style, fluent in form, masters of mimicry. But they are NOT reliable knowledge engines or databases. If you build products on top of them, you are betting your future on a storyteller that never knows when it's lying. I feel like Cato the Elder since I always end with stuff like this: "Furthermore, I think we need to remember that 'AI' is not only referring to GPT-based LLMs."
-
Niki Athanasiadou, PhD reacted on thisNiki Athanasiadou, PhD reacted on this💜 🎓 Excited to welcome Yann LeCun as our 2026 Tandon Commencement speaker! #NYUTandonMade
-
Niki Athanasiadou, PhD reacted on thisNiki Athanasiadou, PhD reacted on thisGreece… a country where things don’t always work as they should. Where systems can feel incomplete, and where there is still so much left to fix, and reimagine. And yet… it is impossible not to be moved by it. Within just two weeks, you can travel through landscapes that feel like entirely different worlds — from mountains and forests, to sunlit islands and wild, untouched coastlines. The diversity is not just geographical; it’s emotional, cultural, and deeply human. Greek wildlife alone is a reminder of how unique this place is. There’s a richness here that often goes unnoticed — an ecosystem as complex and layered as the country itself. For me, Greece is more than a home. It’s where I return to think, to reset, and to find inspiration before stepping into a new chapter. It grounds me, challenges me, and somehow always manages to surprise me. An unfinished country, yes.But also an endlessly inspiring one.
-
Niki Athanasiadou, PhD reacted on thisDr. Rumman Chowdhury on why she believes AI Safety needs to be mandated, validated and rigorous! ⬇️Niki Athanasiadou, PhD reacted on thisThere is a foundational problem with how the AI industry evaluates the safety of its own systems, and it deserves to be named plainly: the methods being used do not meet the standards we would require in any other domain of consequential engineering. Benchmarks — the primary instrument of AI safety evaluation — are question-and-answer pairs assembled by the same parties who have a material interest in favorable outcomes. Red teaming, while genuinely useful as a discovery practice, is not a replicable, systematic methodology; it is closer to structured improvisation. The result is an evaluation ecosystem that is simultaneously expensive, performative, and epistemically weak. This is not a problem that voluntary industry commitment can resolve, because the structural incentive to produce favorable results does not disappear when participation is optional. Safety evaluation needs to be externally mandated and methodologically rigorous — the same standard we apply to pharmaceuticals, to aviation, to civil infrastructure. During the AI Summit in India, I discussed this in detail with the Hindustan Times, and the full interview is linked below. https://lnkd.in/gtU8VhmtWorld lacks rigorous scientific methods to test AI safety: Rumman ChowdhuryWorld lacks rigorous scientific methods to test AI safety: Rumman Chowdhury
-
Niki Athanasiadou, PhD reacted on thisNiki Athanasiadou, PhD reacted on thisEveryone's running towards AI Engineering. Meanwhile, I feel Data Science roles are a big miss. All the traditional problems still need solving: → Regression → Classification → Time Series → Demand Forecasting, etc Across different domains - Marketing, Customer Analytics, etc. Data Science is not prompt engineering or calling a LLM API. It's about: → Understanding the domain and business problem deeply → Cleaning, wrangling, interpreting raw data → Asking the right questions before modeling → Building groundtruth & robust statistical foundations → Designing experiments, not just tuning parameters → Translating insights into business impact → Making the models explainable to support business, etc. I understand that now Data Scientists need to move beyond notebooks. You will definitely not regret learning and building e2e ML projects. Working on both side, I can definitely say: There is a lot of work to be done on both sides! And you really need to understand fundamentals here, it’s built on fundamentals and your ability to explain them. Both paths lead to great careers. Build depth, not just breadth. Pick the domain that excites you. Don't just go with the Hype. True Data Science is here to stay! Do you agree? -- ♻️ Repost if you agree 💚 ➕ Join 45.500+ serious AI/ML builders here: https://lnkd.in/ds_SzEUH
-
Niki Athanasiadou, PhD reacted on thisNiki Athanasiadou, PhD reacted on thisI'm thrilled to share my last day calendar to motivate you in your PhD 🫰🏻🫰🏻🫰🏻
-
Niki Athanasiadou, PhD liked thisNiki Athanasiadou, PhD liked thisSave the date! https://lnkd.in/dJ4pWmFh 2026, the year to celebrate 30 years of Olia being a net artist! Near or far, join me for the opening on Saturday, June 27 19:00 at City Gallery, Sindelfingen
-
Niki Athanasiadou, PhD reacted on thisNiki Athanasiadou, PhD reacted on thisWe're on the cover of this week's Nature! We propose a model of how cells learn by running an evolutionary algorithm: exploring different gene-regulatory combinations and using feedback responses to stabilize those combinations that reduce stress levels. Full text: https://lnkd.in/ewbZ82tJ
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
-
-
-
-
-
-
Publications
-
Please visit the following site for a complete list of my publications
-
https://www.ncbi.nlm.nih.gov/myncbi/browse/collection/54606748/?sort=&direction=
Honors & Awards
-
Data Science 1st place award
NYC Open Data Project Gallery Contest
-
Promega Young Biochemist Award
Biochemical Society, UK
-
Darwin Trust Scholarship
Edinburgh, UK
-
Lilian Voudouri Foundation Scholarship
Athens, Greece
-
Academic Achievement Award
State Scholarship Foundation, Greece
Languages
-
R
Full professional proficiency
-
bash
Limited working proficiency
-
LaTeX
Limited working proficiency
-
Python
Professional working proficiency
-
MySQL
Limited working proficiency
Other similar profiles
Explore more posts
-
Dr. Lucas Schütz
"Data available on reasonable request" is a phrase that belongs in the past. 🛑 Acquiring spatial biology datasets easily costs tens of thousands of dollars, with large cohorts routinely reaching the $100k mark. Because these efforts are (at least in academia) usually paid for out of public funds, making these datasets genuinely accessible shouldn't be optional, it’s a necessity. But how do we make them available? Classic image databases often aren't built to handle the complexity of spatial proteomics and transcriptomics. Even when previews are available, they are rudimentary at best and strip away the rich context. We need better infrastructure across the industry / community and work together to tackle this challenge. Data hosting needs to be intuitive, accessible, and allow researchers to explore the downstream analysis that has already been done on the dataset, not just struggle with raw, unformatted files. What do you think about this? #FAIRdata #Spatialomics #Openscience
1 Comment -
Scripps Research
Scripps Research scientists Andrew Su and Chunlei Wu have published a new mSystems article describing the creation of the National Institute of Allergy and Infectious Diseases (NIAID) Discovery Portal, a metadata-driven platform that enables discovery of infectious and immune-mediated disease datasets across multiple repositories. The work highlights how harmonized metadata and user-centered design can improve data findability and reuse in biomedical research. Read ‘The NIAID Discovery Portal: a unified search engine for infectious and immune-mediated disease datasets’: https://ow.ly/n1XV50XSzYK #BiomedicalData #DataDiscovery #OpenScience #FAIRData #InfectiousDisease
-
Dan Ryder
Good morning everyone! Check out our latest blog - Bioinformatics in GLP-1 and metabolic drug development - where we share how it helps cut through the noise and accelerate discovery. Remember us for flexible bioinformatic staff augmentation in drug discovery! #GLP1 #Bioinformatics #MetabolicDisease #StaffAugmentation #DrugDevelopment
-
Michelle Holko
New opinion piece about the strategic importance of biological data with John Wilbanks and Sam Howell. One point I want to emphasize: progress here isn’t just about collecting more biological data. To make AI truly transformative for biotechnology, we need advances in data diversity, quality, interoperability, and security—together. If any one of these lags, the entire ecosystem becomes less effective. Building this foundation will require coordinated effort across research, industry, and government. Take a look and I’d welcome your thoughts. 🔗 https://lnkd.in/dgSZ68Gs
8 Comments -
Marc Malandro
The Chan Zuckerberg Initiative has released precomputed TranscriptFormer embeddings for the entire CZ CELL×GENE Census — covering 106M human cells and 42M mouse cells, spanning 1000+ cell types in healthy tissue and across hundreds of disease contexts. #czi #czscience #openscience #aiinscience #transcriptformer
3 Comments -
Gary Shiels
The first in a series of posts focusing on the application of foundation model techniques to epigenetic datasets, providing a variety of perspectives on this area of research, including overviews of major models, potential real-world applications, worked examples of fine-tuning, and deep dives into technical underpinnings. Please have a read and let us know your thoughts. https://bit.ly/3Z42Qbo
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content