Review notes for SNAME

# Review notes for SNAME - Meeting on Monday. 9 AM PST. Major issues - [x] FA writes one ore more paragraphs about other ML algorithms - [ ] BP writes a response to reviewer - [ ] LH to fix minor issues. to fix references  for reviewer 2. - [ ] FA address (1) Add an overview reference. Mention the other areas that affects by ML. - [ ] BP address (2) - [ ] LH will double check both  ## Fredrik (200625) I have added according to the checklist below. The document is uploaded to the Dropbox-folder. What needs to be done: - References need to be updated. I have put them in "comments" in Word. They are also found below. - There are many small editorial changes that needs to be looked over. I have left these out as I guess it's easy for you to fix. For discussion: - Which Figures to keep or remove? - Do we want to add anything more? (I have not addressed all comments, this can be discussed) ----- ## Reviewer one Fredrik: - [x] Q-learning reinforcement learning optimisation approach - [x] Principal Component Analysis (PCA) - [x] Gaussian processes (GP) - [x] Least Absolute Shrinkage and Selector Operator - [x] Support Vector Machines ### Major issues - [x] explain the principles of a Neural Network, and also, @frahlg - [x] Deep Neural Network (DNN) - [x] the very basic principles of other Machine Learning techniques that are mentioned throughout the manuscript. @frahlg - [x] Check which algorithms are mentioned and make short description of each. @frahlg - [x] linear regression @frahlg - [x] I recommend adding a figure with a concept scheme that shows the different types of Machine Learning techniques and applications in the shipping industry. - I guess this could be done with PP, or maybe some simple flow chart-program. - [x] Also, adding new figures to summarise the concepts introduced in the manuscript would help to enhance the presentation of your manuscript. - Scikit-learn cheat-sheet... ### Minor issues - [ ] In page 2, complete reference information is missing for Brynjolfsson and Mitchell. - [ ] In page 2, I recommend adding a reference for the sentence ‘Depending on how the learning task is achieved, machine learning algorithms can be classified into Supervised Learning, Unsupervised Learning, Semi-supervised Learning and Reinforcement Learning’. - [x] Figures must have a caption starting with the abbreviated name, i.e. Fig. 1. Please apply this format for all the figures in your manuscript. - [ ] Please, refer to figure 1 and figure 2 within the main text of the manuscript. In the current version of your paper, there is no obvious contribution of both figures. - I suggest that we remove the Greentech figure. It's still there, and I have moved the text, but I am not sure which value it provides. - - [x] Regarding the term ‘Neural Network’, I recommend using ‘Artificial Neural Network’. - I have added ANN to the description. - [x] Regarding the term ‘Gaussian Process (GP)’, I recommend using ‘Gaussian Process model (GP)’. - [x] In page 3, second paragraph, I recommend defining ‘Neural Networks’. - [x] In page 3, second column, the authors mention ‘Deep Neural Network’. Could you please explain the difference between ‘Neural Network’ and ‘Deep Neural Network’. - Explained in the part of ANN, ML fundamentals. - [ ] Regarding the reference Peterson et al. (2012), please confirm that the information is correct in the text and in the references list. I understand that the correct spelling of the name should be Petersen. - [ ] In page 2, one of the conditions for implementing Machine Learning is ‘The phenomenon or function being learned should not change rapidly over time’, whereas in page 4 it is stated the Ship Performance Model (SPM) can be divided in (i) a static part and (ii) a dynamic part. Also, it is said that the algorithm by Liu and Bucknall (2015) works for dynamic environments with other moving vessels. For my opinion, saying that the function being learned should not change rapidly over time, and considering a dynamic part of a model, are somehow contradictory. Could you please discuss this apparent conflict in the manuscript? - [ ] I recommend you reviewing, and citing if suitable for your manuscript, the following references: - Aleksandar-Saša Milaković, Fang Li, Mohamed Marouf & Sören Ehlers (2019) A machine learning-based method for simulation of ship speed profile in a complex ice field, Ships and Offshore Structures. https://doi.org/10.1080/17445302.2019.1697075 - Chen Cheng, Peng-Fei Xua, Hongxia Cheng, Yanxu Ding, Jinhai Zheng, Tong Ge, Dianhong Sun & Jin Xu. Ensemble learning approach based on stacking for unmanned surface vehicle's dynamics. https://doi.org/10.1016/j.oceaneng.2020.107388 - Yiannis Raptodimos & Iraklis Lazakis (2018) Using artificial neural network-self-organising map for data clustering of marine engine condition monitoring applications, Ships and Offshore Structures, 13:6, 649-656. https://doi.org/10.1080/17445302.2018.1443694 - Yu, S., Wang, L., Li, B. et al. Optimal setpoint learning of a thruster-assisted position mooring system using a deep deterministic policy gradient approach. J Mar Sci Technol (2019). https://doi.org/10.1007/s00773-019-00678-5 ### Add the following to the references: - [ ] Christos Gkerekos, Iraklis Lazakisa & GerasimosTheotokatos (2019). Machine learning models for predicting ship main engine Fuel Oil Consumption: A comparative study, Ocean Engineering, 15, 106282. https://doi.org/10.1016/j.oceaneng.2019.106282 - [ ] F. Ahlgren, M. E. Mondejar, and M. Thern, ‘Predicting Dynamic Fuel Oil Consumption on Ships with Automated Machine Learning’, Energy Procedia, vol. 158, pp. 6126–6131, Feb. 2019, doi: 10.1016/j.egypro.2019.01.499 - [ ] Scikit-learn: Machine Learning in Python, Pedregosa et al., JMLR 12, pp. 2825-2830, 2011. - [ ] Python Machine Learning Second Edition. Sebastian Raschka and Vahid Mirjalili (2017) Packt Publishing - [ ] Randal S. Olson, Ryan J. Urbanowicz, Peter C. Andrews, Nicole A. Lavender, La Creis Kidd, and Jason H. Moore (2016). Automating biomedical data science through tree-based pipeline optimization. Applications of Evolutionary Computation, pages 123-137. - [ ] Auto-WEKA 2.0: Automatic model selection and hyperparameter optimization in WEKA Lars Kotthoff, Chris Thornton, Holger Hoos, Frank Hutter, and Kevin Leyton-Brown. In JMLR. 18(25):1−5, 2017 - [ ] Transport 2040: Automation, Technology, Employment - The Future of Work (2019) [link](https://commons.wmu.se/cgi/viewcontent.cgi?article=1071&context=lib_reports) - [ ] Reinforcement Learning: An Introduction second edition Richard S. Sutton and Andrew G. Barto The MIT Press Cambridge, Massachusetts - [ ] An Introduction to Statistical Learning with Applications in R. Gareth James • Daniela Witten • Trevor Hastie Robert Tibshirani. Springer - [ ] Kevin P. Murphy. Machine Learning a Probabilistic Perspective (2012, The MIT Press) ## Reviewer 2 Comments to the Author This paper is well written and in good quality. I do not see technical deficiencies and therefore recommend being published with the following comments. - Besides the three areas discussed in this paper, there are several other notable areas in marine industry where ML is widely considered as part of the solutions (may or may not be for operational efficiency), such as asset reliability and condition monitoring, navigation assistance, such as obstacle identification, autonomous operations and autonomous navigation. As a review paper, the author may consider add a short paragraph to brief the other potential areas for the scholars to have an overall view on the potential opportunities. - [ ] Fredrik. Add an overview reference. Mention the other areas that ---- There are several other areas in shipping that are impacted by machine learning and not discussed in this paper, such as autonomous ships, condition monitoring and maitenance. Ports can leverage on ML for real time data from cargo containers, and also minimising the manual work of paperwork when it goes digital. Also, not only the fuel consumption but also the emissions and underwater noise can be reduced, by combining data from en engine performance and routes. ---- - The second paragraph in DISCUSSION section, the author discusses the possible reason that traditional knowledge still dominates the maritime industry. The author mainly discusses the data availability and communication and computational power limitations although with mentioning “creating, optimizing and maintaining relevant algorithms will require expertise and extensive human inputs”. Given that many applications in marine industry requiring reliable, accurate and conservative outcomes since the potential consequence of errors are significant. As the author has pointed out in MACHINE LEARNING FUNDAMENTALS, ML can be implemented for “The task has a tolerance for error and no need for provably correct or optimal solutions”. For the above reason, I believe human inputs, guidance, and verification will be a critical piece, along with the implementation of ML in the marine industry for the foreseeable future. If the author accepts the above statement, I suggest the author to reword this paragraph to reflect the statement. - The third paragraph in DISCUSSION section, “The industry should be aware that such a scenario would cause corresponding job cuts”. I believe in addition to the possible job cuts, a working force transformation from traditional roles to a new role who is equipped with advanced knowledge, such as ML, is worthwhile to be mentioned. - [x] Mention new roles in the working force. Perhaps, 2019 Transport 2040: Automation, Technology, Employment - The Future of Work [link](https://commons.wmu.se/cgi/viewcontent.cgi?article=1071&context=lib_reports)