The crawl is chuckflow i.e. SQS. The mask is BossDB.
The crawl is sequenced by the FFN algorithm deciding where to go next as it already does, this is just a different substrate for queueing up explorations.
First FFN paper Figure 1 has "image channel" and "mask input channel" for 33x33x17 cuboids. Image is single channel. Could represent data as one cuboid of 2-channel data. But the idea of two cuboids in sounds more powerful. Image data is in repo. Mask data is also stored in repo, but separate. One is truth input data, other is a run of a program.
The crawl is chuckflow i.e. SQS. The mask is BossDB.
The crawl is sequenced by the FFN algorithm deciding where to go next as it already does, this is just a different substrate for queueing up explorations.
First FFN paper Figure 1 has "image channel" and "mask input channel" for 33x33x17 cuboids. Image is single channel. Could represent data as one cuboid of 2-channel data. But the idea of two cuboids in sounds more powerful. Image data is in repo. Mask data is also stored in repo, but separate. One is truth input data, other is a run of a program.