The topic of the talk is: Legal Document Summarization and AI Explainability
Abstract: “Legal research is a highly manual and time-consuming process. Legal professionals, for example, have to read court cases that are up to 100 pages long, just to identify the most important aspects in order to decide whether their firm should represent the case.
Natural Language Processing (NLP) techniques like information extraction and summarisation thus provide great opportunities to save law firms time. However, applying NLP to legal documents is highly challenging due to the domain-specific terminology and variability in the legal document layout.
In this talk, we show how Thomson Reuters Labs tackles these challenges and present our work on summarising court cases and providing human-understandable explanations thereof.”
The meeting is open to all interested – feel free to join!
The topic of the talk is: AI Governance, Standards and Regulation for a Trustworthy AI ecosystem
Abstract: “The increasing use of algorithmic decision-making technologies (colloquially referred to as “AI”) in industry, commerce and public service provision are giving rise to concerns about potential negative impacts on individuals (e.g. algorithmic discrimination bias) and the wider socio-economic fabric of society (e.g. displacement of jobs).
With trust in the technology being cited as a key barrier for successful deployment, both in the public and private sector, there is a growing push toward translating AI ethics principles into actionable practice.
This talk will review current initiatives towards the development of AI governance frameworks, ethics and oversight related standards and regulations, and discuss the role each of these can play within a wider ecosystem of trustworthiness for the use of AI.”
The meeting is open to all interested – feel free to join!
The topic of his talk is: Predicting the Unpredictable: Mathematical Models and the Covid 19 Pandemic
Abstract of the presentation: “Predicting the future of the COVID-19 pandemic is a challenging task and can not be done without mathematical models describing the progression of the epidemic.
Despite large uncertainties about epidemiological relevant medical and social key parameters, mathematical models can provide deep insights into the dependency of the epidemic dynamics on those parameters.
Epidemiological models can furthermore be used to develop and improve rational strategies for controlling the COVID-19 epidemics.
We focus in the talk on fundamental mathematical features of individual based epidemic models and highlight the close relation to problems in percolation and random graph theory.
We emphasize the special role of households and discuss some results and conjectures in first passage percolation and their impact on epidemic processes.
Finally we present some outcomes of the MOCOS microsimulation model for the COVID -19 epidemic in Poland and Germany and reflect on problems related to model based policy advice for epidemic control.”
The meeting had a form of mini-course focused on Recurrent Neural Networks. The presenter was Karol Draszawka
Karol’s description of course:
“This course presents the basics of Recurrent Neural Networks (RNNs) in the context of sequence labelling problems. The need for RNNs is shown first. After that, basic RNNs architecture and the training algorithm Backpropagation Through Time (BPTT) are given in theory and implemented using Tensorflow primitives. Considatations about problems with such basic RNNs leads to the design of refined RNN architectures, such as LSTM and GRU, which are next implemented, also from Tensorflow primitives. Lastly, some examples and tips for using RNNs for real problems are given.”
Our next meeting is on 9th July 2018. All ML Enthusiasts are welcome to join us!
Łukasz Czekaj gave lecture about Bayesian modelling. He described thoroughly the concept and the most popular approaches (Bayesian networks, Bayesian hierarchical models). Łukasz also presented how to use JAGS framework to build and run those kinds of models.
Our next meeting will take place in two weeks. Feel free to join us!
The 20th meeting of our group took place on 22th of May.
The topic of the meeting was Support Vector Machines. It was thoroughly described by Krzysztof Czarnowski. The presentation was followed by interesting discussion about some mathematical details of those algorithms.
On 24th of April 2017 we had 19th meeting of our group.
Two topics were presented. First presenter was Maciej Godek, who talked about logic programming in Scheme. He also showed us some programs he created – neural networks and genetic algorithm implementations. Additionally, Maciej presented the framework for creating virtual robots.
Second presentation was delivered by Kuba Domaszewicz. The topic of this talk was ECG signal analysis. Methods described by Kuba are used in aidlab device (health tracking wearable assistant).
Next meeting will be held on 8th of May. Feel free to join us!
This time we had opportunity to listen to two speakers. The first – great – presentation was presented by Karol Draszawka, who with extraordinary diligence and enthusiasm discussed the intricacies of AlphaGo algorithm. Thanks to Karol, we learnt about the details and structure of five neural networks created by Google to defeat the masters of Go. The presentation was based on the article “Mastering the game of Go with deep neural networks and tree search“.
The second speaker was Piotr Orzechowski from Trineo company from Gdynia. He gave us an overview of the tasks for which image recognition algorithms are used in the security domain.
Our next meeting will be held on 24th of April 2017. See you!
15th meeting of MLGdańsk group was held on 3rd March at The Gdansk University of Technology.
This time, Marcin Zadroga presented Dataiku Data Science Studio, which is a simple and user friendly tool for data analysis. Dataiku DSS out-of-the-box offers basic and more sophisticated algorithms such as Random Forrest, Logistic Regression, Decision Trees and other.
Marcin showed us how to perform common tasks like data cleaning, training of prognostic model and its validation and evaluation.
Next meeting is on 20th March, you are welcome to join us!
On 20th February we met at The Gdańsk University of Technology for 14th meeting of MLGdańsk group.
Adam Brzeski carried out the second part of the Transfer Learning workshop. We learnt how to retrain neural network model according to transfer learning paradigm. Using Inception model we were able to train it and use it for the image classification tasks.
Our next meeting is on 6th March. All ML Enthusiasts are welcome to join us!
The 13th meeting of our group took place on 6th February.
Adam Karwan talked about his experience with MBA program he participates in – MBA Innovation and Data Analysis. Adam described overall program of studies and the most interesting modules.
Second part of the meeting was dedicated to the transfer learning workshop. This part was led by Adam Brzeski. He presented the paridigm of the transfer learning and showed us how to use TensorFlow for this purpose. He outlined thoroughly the idea of image recognition with Convolutional Neural Networks.
Our next meeting and the second part of the workshop will be held on 20nd February. Feel free to join us!