Machine learning and data analysis: we solve practical problems with the winners of the industrial hackathon LK

How to calculate the intentions of cybercriminals attacking an industrial facility and recognize the weak SOS signals that the industrial process control system emits periodically against the background of “normal” behavior - we will talk about this and much more this coming Wednesday, June 7, at the CoLaboratory: Deep Learning meeting in central Kaspersky Lab’s office. All those who are not indifferent to the topic of industrial safety will have an exciting immersion in the world of machine learning and data analysis under the guidance of the winners of the spring industrial hackathon of the LC and experts of our company.

The heroes of the evening will talk about how to fight the huge 44 GB data set of technological signals of the industrial system and emerge victorious after discovering attacks that are almost impossible to identify. Moreover, the public will be presented with two solutions to this problem - from the participants of the hackathon and from its compilers.

So, the winners of the hackathon will talk about how:

• turn the finding of abnormal behavior in the logs of factory devices into a task of ordinary classification;
• how to generate light but useful features that characterize the entire time series in a short time;
• how to divide the work within the team;
• how to "mix" different models on one set of features;
• how to raise AUC with kNN cherries on Competitive Data Science cake.

Small spoiler: the approach of our heroes is based on the "three pillars":

1. a key decision based on counting time series statistics as features;
2. coding consecutive rarefied system states by a feature vector;
3. identifying anomalies using LSTM, ideas for their modifications.

Well, experts from Kaspersky Lab will share their vision of the problem of data analysis of ACU TP: what criteria they were guided by when compiling tasks and a data set for participants, what they expected from the teams and what happened in the end.

In short, on June 7 at 19.00 we are waiting for everyone to practice at Deep Machine Learning at the headquarters of the LC. You can register for the event here .