![\"enter](\"https://si.wsj.net/public/resources/images/BN-RF681_1215CS_P_20161214214245.jpg\" "\\"\\"")
It could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"
\n\n
http://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066
", "messages": [ { "id": "150544", "postDate": "12/15/2016 17:40:04", "content": "It could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"
\n\n
http://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066
", "rawMarkdown": "It could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"\r\n\r\n![enter image description here][1]\r\n\r\nhttp://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\r\n\r\n\r\n [1]: https://si.wsj.net/public/resources/images/BN-RF681_1215CS_P_20161214214245.jpg", "votes": null }, { "id": "150562", "postDate": "12/15/2016 19:53:41", "content": "[quote=Bojan Tunguz;150544]
\n\nIt could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"
\n\n
http://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\nDoes not macke any logic, only if you are reffering on weapons of mass distruction :))
\n\n[/quote]
", "rawMarkdown": "[quote=Bojan Tunguz;150544]\r\n\r\nIt could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"\r\n\r\n![enter image description here][1]\r\n\r\nhttp://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\r\nDoes not macke any logic, only if you are reffering on weapons of mass distruction :))\r\n\r\n [1]: https://si.wsj.net/public/resources/images/BN-RF681_1215CS_P_20161214214245.jpg\r\n\r\n[/quote]", "votes": null }, { "id": "150579", "postDate": "12/15/2016 21:25:12", "content": "Bojan, I don't get your point. Almost every technology we have has a \"dual use\" if you mean that it can be used for civil or military uses. The \"d\" in dstl stands for defence, which of course will have military applications for any machine learning or statistical methods developed in this competition. But then, everyone else in the world with knowledge of how to apply them has access to these methods, which in my mind, balances the playing field somewhat.
\n\nIn the interests of peaceful coexistence on this planet, since that is what seems to be implied by your post, I would like to see people from all nations and political ideologies contribute to this project for both civil and defence purposes. The more evenly distributed the knowledge the better, imo.
", "rawMarkdown": "Bojan, I don't get your point. Almost every technology we have has a \"dual use\" if you mean that it can be used for civil or military uses. The \"d\" in dstl stands for defence, which of course will have military applications for any machine learning or statistical methods developed in this competition. But then, everyone else in the world with knowledge of how to apply them has access to these methods, which in my mind, balances the playing field somewhat.\r\n\r\nIn the interests of peaceful coexistence on this planet, since that is what seems to be implied by your post, I would like to see people from all nations and political ideologies contribute to this project for both civil and defence purposes. The more evenly distributed the knowledge the better, imo.", "votes": null }, { "id": "150580", "postDate": "12/15/2016 21:30:08", "content": "[quote=r3m;150579]
\n\nBojan, I don't get your point.
\n\n[/quote]
\n\nI was not trying to make a point, at least not a very contentious one. Just saw that article come up on top of my news feed today, and it immediately resonated with this challenge. It struck me that this was exactly the kind of stuff that a military-connected organization would be looking for, and the it's very likely that the algorithms developed here will end up being used for such purposes. That's all.
", "rawMarkdown": "[quote=r3m;150579]\r\n\r\nBojan, I don't get your point.\r\n\r\n[/quote]\r\n\r\nI was not trying to make a point, at least not a very contentious one. Just saw that article come up on top of my news feed today, and it immediately resonated with this challenge. It struck me that this was exactly the kind of stuff that a military-connected organization would be looking for, and the it's very likely that the algorithms developed here will end up being used for such purposes. That's all.", "votes": null }, { "id": "150586", "postDate": "12/15/2016 21:59:42", "content": "Oh, I see. You were sharing a recent discovery. Yes, I just take it for granted that one affects the other.
\n\nOf course there are problems with the use of machine learning for defence purposes, not the least of which are adversarial structures generated for the purpose of falsely identifying weapons locations. These structures could be physical or virtual (at the image level) and would contribute to a higher variance in identification of weapons locations, ultimately making a tool such as this unreliable as a sole means of identification. The defence people of course know these shortcomings, so, practically speaking, the application of the results of this contest would favour civil over defence uses.
\n\nNow, what would be very interesting to me is if this satellite data had some adversarial structures encoded in them which forced the development of methods which rejected them. Then those methods which utilise adversarial gradient training, or similar, should show better results.
", "rawMarkdown": "Oh, I see. You were sharing a recent discovery. Yes, I just take it for granted that one affects the other.\r\n\r\nOf course there are problems with the use of machine learning for defence purposes, not the least of which are adversarial structures generated for the purpose of falsely identifying weapons locations. These structures could be physical or virtual (at the image level) and would contribute to a higher variance in identification of weapons locations, ultimately making a tool such as this unreliable as a sole means of identification. The defence people of course know these shortcomings, so, practically speaking, the application of the results of this contest would favour civil over defence uses.\r\n\r\nNow, what would be very interesting to me is if this satellite data had some adversarial structures encoded in them which forced the development of methods which rejected them. Then those methods which utilise adversarial gradient training, or similar, should show better results.", "votes": null } ], "comments": [ { "id": 150562, "author_name": "", "author_url": "", "post_date": "12/15/2016 19:53:41", "content": "[quote=Bojan Tunguz;150544]
\n\nIt could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"
\n\n
http://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\nDoes not macke any logic, only if you are reffering on weapons of mass distruction :))
\n\n[/quote]
", "votes": null, "replies": [] }, { "id": 150579, "author_name": "rshpeley", "author_url": "", "post_date": "12/15/2016 21:25:12", "content": "Bojan, I don't get your point. Almost every technology we have has a \"dual use\" if you mean that it can be used for civil or military uses. The \"d\" in dstl stands for defence, which of course will have military applications for any machine learning or statistical methods developed in this competition. But then, everyone else in the world with knowledge of how to apply them has access to these methods, which in my mind, balances the playing field somewhat.
\n\nIn the interests of peaceful coexistence on this planet, since that is what seems to be implied by your post, I would like to see people from all nations and political ideologies contribute to this project for both civil and defence purposes. The more evenly distributed the knowledge the better, imo.
", "votes": null, "replies": [ { "id": 150580, "author_name": "tunguz", "author_url": "", "post_date": "12/15/2016 21:30:08", "content": "[quote=r3m;150579]
\n\nBojan, I don't get your point.
\n\n[/quote]
\n\nI was not trying to make a point, at least not a very contentious one. Just saw that article come up on top of my news feed today, and it immediately resonated with this challenge. It struck me that this was exactly the kind of stuff that a military-connected organization would be looking for, and the it's very likely that the algorithms developed here will end up being used for such purposes. That's all.
", "votes": null, "replies": [] } ] }, { "id": 150586, "author_name": "rshpeley", "author_url": "", "post_date": "12/15/2016 21:59:42", "content": "Oh, I see. You were sharing a recent discovery. Yes, I just take it for granted that one affects the other.
\n\nOf course there are problems with the use of machine learning for defence purposes, not the least of which are adversarial structures generated for the purpose of falsely identifying weapons locations. These structures could be physical or virtual (at the image level) and would contribute to a higher variance in identification of weapons locations, ultimately making a tool such as this unreliable as a sole means of identification. The defence people of course know these shortcomings, so, practically speaking, the application of the results of this contest would favour civil over defence uses.
\n\nNow, what would be very interesting to me is if this satellite data had some adversarial structures encoded in them which forced the development of methods which rejected them. Then those methods which utilise adversarial gradient training, or similar, should show better results.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "150544": "It could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"\r\n\r\n![enter image description here][1]\r\n\r\nhttp://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\r\n\r\n\r\n [1]: https://si.wsj.net/public/resources/images/BN-RF681_1215CS_P_20161214214245.jpg", "150562": "[quote=Bojan Tunguz;150544]\r\n\r\nIt could come in really handy. On the other hand, the algorithms we develop here could potentially be classified (no pun intended!) as \"dual-use technology.\"\r\n\r\n![enter image description here][1]\r\n\r\nhttp://www.wsj.com/articles/china-installs-weapons-in-south-china-sea-satellites-show-1481771066\r\nDoes not macke any logic, only if you are reffering on weapons of mass distruction :))\r\n\r\n [1]: https://si.wsj.net/public/resources/images/BN-RF681_1215CS_P_20161214214245.jpg\r\n\r\n[/quote]", "150579": "Bojan, I don't get your point. Almost every technology we have has a \"dual use\" if you mean that it can be used for civil or military uses. The \"d\" in dstl stands for defence, which of course will have military applications for any machine learning or statistical methods developed in this competition. But then, everyone else in the world with knowledge of how to apply them has access to these methods, which in my mind, balances the playing field somewhat.\r\n\r\nIn the interests of peaceful coexistence on this planet, since that is what seems to be implied by your post, I would like to see people from all nations and political ideologies contribute to this project for both civil and defence purposes. The more evenly distributed the knowledge the better, imo.", "150580": "[quote=r3m;150579]\r\n\r\nBojan, I don't get your point.\r\n\r\n[/quote]\r\n\r\nI was not trying to make a point, at least not a very contentious one. Just saw that article come up on top of my news feed today, and it immediately resonated with this challenge. It struck me that this was exactly the kind of stuff that a military-connected organization would be looking for, and the it's very likely that the algorithms developed here will end up being used for such purposes. That's all.", "150586": "Oh, I see. You were sharing a recent discovery. Yes, I just take it for granted that one affects the other.\r\n\r\nOf course there are problems with the use of machine learning for defence purposes, not the least of which are adversarial structures generated for the purpose of falsely identifying weapons locations. These structures could be physical or virtual (at the image level) and would contribute to a higher variance in identification of weapons locations, ultimately making a tool such as this unreliable as a sole means of identification. The defence people of course know these shortcomings, so, practically speaking, the application of the results of this contest would favour civil over defence uses.\r\n\r\nNow, what would be very interesting to me is if this satellite data had some adversarial structures encoded in them which forced the development of methods which rejected them. Then those methods which utilise adversarial gradient training, or similar, should show better results." }, "source": "meta" }