Differences between revisions 739 and 807 (spanning 68 versions)
|
Size: 4807
Comment:
|
← Revision 807 as of 2026-08-09 08:57:46 ⇥
Size: 2662
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 3: | Line 3: |
| = Natural Language Processing Seminar 2025–2026 = | = Natural Language Processing Seminar 2026–2027 = |
| Line 7: | Line 7: |
| ||<style="border:0;padding-top:5px;padding-bottom:5px">'''15 September 2025'''|| ||<style="border:0;padding-left:30px;padding-bottom:0px">'''Louis Esteve''' (Universite Paris-Saclay) || ||<style="border:0;padding-left:30px;padding-bottom:5px">[[https://zil.ipipan.waw.pl/seminarium-online|{{attachment:seminarium-archiwum/teams.png}}]] '''Diversity and dataset size – a quantitative perspective'''  {{attachment:seminarium-archiwum/icon-en.gif|Talk in English.}}|| ||<style="border:0;padding-left:30px;padding-bottom:15px">The field of Natural Language Processing (NLP) studies the abilities of computer systems to process and generate natural language, and has received increasing attention from the general population since the democratisation of generative and conversational models. However, behind the scenes, state-of-the-art NLP models are trained on ever-larger datasets, reaching trillions of tokens. It may be argued that the creation and use of such immense datasets is motivated by the idea that 'the larger the dataset, the more diverse it is', and that in turn 'if the training set is more diverse, it shall yield better models'. However, these statements thus far remain intuitions and need to be properly tested. To this end, this presentation will tackle methods and caveats of formal diversity quantification including limitations of the literature, a preliminary discussion on the link between diversity and dataset size, as well as their impact on downstream applications.|| ||<style="border:0;padding-top:10px">Please see also [[http://nlp.ipipan.waw.pl/NLP-SEMINAR/previous-e.html|the talks given in 2000–2015]] and [[http://zil.ipipan.waw.pl/seminar-archive|2015–2025]].|| |
||<style="border:0;padding-top:10px">Please see also [[http://nlp.ipipan.waw.pl/NLP-SEMINAR/previous-e.html|the talks given in 2000–2015]] and [[http://zil.ipipan.waw.pl/seminar-archive|2015–2026]].|| |
| Line 15: | Line 10: |
||<style="border:0;padding-top:5px;padding-bottom:5px">'''17 November 2025''' '''(NOTE: the seminar will start at 16:00)'''|| ||<style="border:0;padding-left:30px;padding-bottom:0px">'''Marzena Karpińska''' (Microsoft) || ||<style="border:0;padding-left:30px;padding-bottom:5px">[[http://zil.ipipan.waw.pl/seminarium-online|{{attachment:seminarium-archiwum/teams.png}}]] '''!OneRuler: testing multilingual language models on long contexts'''  {{attachment:seminarium-archiwum/icon-pl.gif|Talk in Polish}}|| ||<style="border:0;padding-left:30px;padding-bottom:15px">In this presentation, I will look at how well language models perform when extracting information from texts of up to 128,000 tokens (approximately 100,000 words) in 26 languages, including Polish. The results of the experiments show that as the length of the context increases, the differences between languages with large and small data resources also increase. Surprisingly, even minimal changes in the command (adding the possibility that the information does not exist) cause a significant decrease in effectiveness, especially with longer texts.|| |
|
| Line 21: | Line 21: |
||<style="border:0;padding-top:5px;padding-bottom:5px">'''2 April 2020'''|| ||<style="border:0;padding-left:30px;padding-bottom:0px">'''Stan Matwin''' (Dalhousie University)|| ||<style="border:0;padding-left:30px;padding-bottom:5px">'''Efficient training of word embeddings with a focus on negative examples'''  {{attachment:seminarium-archiwum/icon-pl.gif|Talk delivered in Polish.}} {{attachment:seminarium-archiwum/icon-en.gif|Slides in English.}}|| ||<style="border:0;padding-left:30px;padding-bottom:15px">This presentation is based on our [[https://pdfs.semanticscholar.org/1f50/db5786913b43f9668f997fc4c97d9cd18730.pdf|AAAI 2018]] and [[https://aaai.org/ojs/index.php/AAAI/article/view/4683|AAAI 2019]] papers on English word embeddings. In particular, we examine the notion of “negative examples”, the unobserved or insignificant word-context co-occurrences, in spectral methods. we provide a new formulation for the word embedding problem by proposing a new intuitive objective function that perfectly justifies the use of negative examples. With the goal of efficient learning of embeddings, we propose a kernel similarity measure for the latent space that can effectively calculate the similarities in high dimensions. Moreover, we propose an approximate alternative to our algorithm using a modified Vantage Point tree and reduce the computational complexity of the algorithm with respect to the number of words in the vocabulary. We have trained various word embedding algorithms on articles of Wikipedia with 2.3 billion tokens and show that our method outperforms the state-of-the-art in most word similarity tasks by a good margin. We will round up our discussion with some general thought s about the use of embeddings in modern NLP.|| |
Natural Language Processing Seminar 2026–2027
The NLP Seminar is organised by the Linguistic Engineering Group at the Institute of Computer Science, Polish Academy of Sciences (ICS PAS). It will restart in October and will take place on (some) Mondays, usually at 10:15 am, often online – please use the link next to the presentation title. All recorded talks are available on YouTube. |
|
Please see also the talks given in 2000–2015 and 2015–2026. |
