RT @paoloceravolo: Tutorial:: Process Mining the Scikit-Learn Way: Introducing SkPM. Sylvio Barbon Junior @raseidi .#ML4PM @icpm_conf https….

RT @paoloceravolo: Presenting CoSMo a Framework for Conditioned Process Simulation at @BPMConf

codebase available as well:

A step towards making deep learning more flexible for the simulation of processes! Very proud of this work. Paper available here

RT @paoloceravolo: Well done @raseidi 🙂

I'm really excited about this! Looking forward to working together and accomplishing great things :)).

Our new paper has been published in the special issue on Artificial Intelligence for Process Mining in the Engineering Applications of Artificial Intelligence journal. Check it out!.

We're always looking to improve and do even better. We value your feedback, so please don't hesitate to contact us!.

By biasing desired outputs, we can mitigate the natural stochasticity of deep neural nets. Our approach unlocks a range of powerful applications, such as synthetic log generation, what-if analysis, and outcome prediction.

We demonstrate the efficacy of our proposal by conditioning the simulations on the resource usage of processes. In practice, the conditioned simulation model can simulate processes relying on the availability of a given resource.

The framework aims to create neural architectures that offer more control and flexibility for process simulations. In other words, users can simulate within predefined constraints.

A new preprint is now available! Designing conditioned networks is the key to successful process simulation models. In this work, we propose CoSMo: a framework for implementing COnditioned process Simulation MOdels!. @gmtavares_ @paoloceravolo.

RT @sbarbonjr: #Metalearning meets Active Learning in Online Machine Learning. In this paper on Elsevier Pattern R….

RT @sbarbonjr: Is one #encoding method better than another? We compared the expressiveness, scalability, correlation power, and domain agno….

Any feedback and suggestions are welcomed. Feel free to get in touch :).

The benchmark focuses on the anomaly detection task and assesses 27 encoding methods through hundreds of event logs. We discuss future directions based on our findings. Therefore, we also expect the proposals and insights presented in our work can be leveraged for other PM tasks.

The metrics cover what we believe is essential for encoding in PM: expressivity of the encoded data, scalability w.r.t. the event log cardinality, correlation between performances of the encoding algorithm performance and the PM task, and domain agnosticism regarding PM tasks.

Lastly, in order to guide researchers and practitioners to choose the right algorithm according to their goals, we proposed new evaluation metrics and performed an extensive experimental evaluation to serve as a benchmark.

Second, we performed a systematic review of PM papers that employ any type of encoding method. Subsequently, we reviewed alternative encoding methods in the literature and classified them into different families. We proposed a new taxonomy to guide this discussion.

First, we show how difficult it is to choose the right encoding method along with its parameters. We also discuss how each algorithm configuration achieves different performances for different event logs.