← Back to Home

Papers i'm reading #2

Continuation from this post.

An unsupervised spatiotemporal graphical modeling approach to anomaly detection in distributed CPS (Cyber Physical Systems).

Link

My Summary: Really interesting paper - PGM’s, HMM’s and all that good stuff. Quite complicated though and no clear route to implementation. Also I would wonder how well it scales beyond 10’s of time series. More good for learning about some different approaches as opposed to anything that could be implemented easily. 

Notes:


Recent Advances in Anomaly Detection Methods applied to Aviation.

Link

My Summary: Interesting paper focusing specifically on aviation but in a broad sense and in an up to date manner covering many newer techniques too. Still seems in reality a lot of domain specific and traditional methods is what is actually used in reality as opposed to in the research.

Notes:


Anomaly Detection in Flight Recorder Data: A Dynamic Data-driven Approach (NASA).

Link

My Summary: A nice look at the different systems and approaches used in aviation. Interesting type of feature engineering proposed Symbolic Dynamic Filtering (SDF). 

Notes:


Histogram-based Outlier Score (HBOS): A fast Unsupervised Anomaly Detection Algorithm

Link

My Summary: Very quick and crisp paper, big focus on computational efficiency and linear time of HBOS. Main downside is HBOS seems mainly univariate. 

Notes:


Visualizing Big Data Outliers through Distributed Aggregation (HDOutliers)

Link

My Summary: Seems to have some nice properties but not clear if is suitable at all for an online setting.

Notes:


Anomaly Detection for Discrete Sequences: A Survey

Link

My Summary: Interesting enough survey of a totally different way of potentially framing time series AD. Paper is from 2012 so a little old, but a good overview of higher level approaches in a more traditional sense.

Notes:


Anomaly Detection in Streams with Extreme Value Theory (SPOT)

Link

My Summary: Very interesting paper, hbos seems to have lots of advantages as being fast and making little assumptions. Very much focused on point anomaly setting.

Notes:


← Back to Home