| 7 Dec 2018 |
 cd10012 | thoughts? | 16:45:14 |
 dOrgJelli | In pandos telegram (hop on over and check it out if you aren’t familiar), they sent these two documents that I think aims to answer this problem (still need to read myself):
https://docs.google.com/document/d/1tYNMLkKpeSbPuPhusqw6LyUCIy08KeZFmv2KrylOPS4/mobilebasic
https://docs.google.com/document/d/1LfJX3dYj-pm1G7bSocEzYlPbT4Pppkc5nkV-lTLAp2U/mobilebasic | 17:06:17 |
| cd10012 | could I get an inv to the pando telegram? | 17:11:06 |
| cd10012 | In reply to @dorgjelli:matrix.org
In pandos telegram (hop on over and check it out if you aren’t familiar), they sent these two documents that I think aims to answer this problem (still need to read myself):
https://docs.google.com/document/d/1tYNMLkKpeSbPuPhusqw6LyUCIy08KeZFmv2KrylOPS4/mobilebasic
https://docs.google.com/document/d/1LfJX3dYj-pm1G7bSocEzYlPbT4Pppkc5nkV-lTLAp2U/mobilebasic Thank you I'll dive into this doc tonight | 17:12:22 |
| cd10012 | gonna also leave this here...https://arxiv.org/pdf/1809.02630.pdf | 21:03:33 |
| cd10012 | but basically this is what i was talking about combining deep learning with semantic AI btw | 21:04:03 |
| cd10012 | and these brilliant folks cracked the nut for the autoencoder to use natural language this is very exciting | 21:05:13 |
| cd10012 | Redacted or Malformed Event | 21:06:22 |
| cd10012 | The difficulty, in part, is owing to the challenge of an efficient parameterization of the graphs while
maintaining semantic validity in context.
| 21:06:55 |
| cd10012 | So for the example they provided they use a grammar to structure their textual data, the goal would be to create a CNF grammar for DAO's that extends the ethereum ontology | 21:19:47 |
| 8 Dec 2018 |
| dOrgJelli | Ahhh I've thought a bit about this with my naive understanding / experience with ML/AI. Is the goal to be able to aggregate all conversation data from the DAO and produce some sort of aggregate meaning(s)? | 03:21:19 |
| cd10012 | Nah so this would be for the encoded data stored as json-ld metadata and having a neural net trained over that data | 04:21:59 |
| cd10012 | so let’s say I have an ontology for daostack and I write an event written with the semantic syntax and store it on the json, we could then train a net on that for optimizing policies to reach goals that we define using that same semantic | 04:24:20 |
| dOrgJelli | yes yes yes yes, all of this | 04:24:48 |
| cd10012 | and this preserves that the semantic prediction given is one of valid structure | 04:25:11 |
| dOrgJelli | yes | 04:25:25 |
| cd10012 | they havent solved the general case tho so I’d have to play with the autoencoder | 04:25:44 |
| cd10012 | for whatever ontology we define and then see if we can put it im chompsky normal form | 04:26:12 |
| dOrgJelli | could you give a simple example of this in terms that aren't technical, but rather practical? | 04:26:25 |
| cd10012 | Sure so let’s say we have a semantics to describe a proposal. So a node on that graph would maybe contain some scalar data and some edges. Edges can be typed so one edge could be for edges to nodes that represent People or the proposal document itself as an entity | 04:31:29 |
| cd10012 | Let’s say I want to have a neural net tell me the probability of Actor A will vote yes on Proposal B. It could then train on a bunch of subgraphs that contain their voting record | 04:35:23 |
| cd10012 | since the data is semantically defined the prediction will say ()% Actor A will vote yes on proposal B based on X, i.e X is the observation that is also semantically defined | 04:37:59 |
| cd10012 | this a toy example because ideally you would want to ask for richer predictions than that but that depends on the semantics. But the magic is that neural net spits out plain english and not just probabilities | 04:40:10 |
| cd10012 | after it goes through the transformation process | 04:40:22 |
| cd10012 | so you’ll need to define more sophisiticated schemas and relations to make more sophisticated features | 04:40:48 |
| dOrgJelli | In the middle of drawing this on my whiteboard and trying to grasp it, thank you so much for this. Will reply with images of what I can comprehend | 04:52:58 |
| dOrgJelli | Oh my god my brain is about to burst, I've been trying to wrap my head around this for some time and this is breaking through the ether, | 04:55:08 |
| dOrgJelli | here's the drawing | 04:55:13 |
| dOrgJelli | one sec, phone's slow | 04:55:51 |
| dOrgJelli | 
Download ima_b0cfb22.jpeg | 04:56:33 |
