Author: Chris Warburton <chriswarbo@gmail.com>
Date: Tue Jan 8 09:09:31 AM UTC 2019
Parent: 2ba29c3c6ee5e7abe74dafaf3cc04d95ad4055cf
Log message:
New issues
1: diff --git a/.issues/ac06e9d2a16b2477/new/1537207932.M386240P2772Q1.nixos b/.issues/ac06e9d2a16b2477/new/1537207932.M386240P2772Q1.nixos 2: new file mode 100644 3: index 0000000..ba213f0 4: --- /dev/null 5: +++ b/.issues/ac06e9d2a16b2477/new/1537207932.M386240P2772Q1.nixos 6: @@ -0,0 +1,15 @@ 7: +From: Chris Warburton 8: +Date: Mon, 17 Sep 2018 19:05:55 +0100 9: +State: new 10: +Subject: Use large, orthogonal vectors 11: +Message-Id: <ac06e9d2a16b2477-0-artemis@nixos> 12: + 13: +Our vectors are quite small, and should also have each of their elements 14: +initialised (pseudo)randomly. Currently we're activating one feature, 15: +but that's a local representation. 16: + 17: +Maybe use floats, initialised with a normal distribution as per: 18: + 19: +Cox, G., Kachergis, G., Recchia, G., & Jones, M. N. (2011). 20: +Towards a scalable holographic word-form representation. 21: +Behavior Research Methods, 43, 602-615. 22: diff --git a/.issues/b41d0b5fcda7c9b4/new/1537222940.M969479P6474Q1.nixos b/.issues/b41d0b5fcda7c9b4/new/1537222940.M969479P6474Q1.nixos 23: new file mode 100644 24: index 0000000..b7b6fac 25: --- /dev/null 26: +++ b/.issues/b41d0b5fcda7c9b4/new/1537222940.M969479P6474Q1.nixos 27: @@ -0,0 +1,16 @@ 28: +From: Chris Warburton 29: +Date: Mon, 17 Sep 2018 23:04:42 +0100 30: +State: new 31: +Subject: Circular convolution is commutative? 32: +Message-Id: <b41d0b5fcda7c9b4-0-artemis@nixos> 33: + 34: +We shouldn't be using a commutative operation. According to: 35: + 36: +Cox, G., Kachergis, G., Recchia, G., & Jones, M. N. (2011). 37: +Towards a scalable holographic word-form representation. 38: +Behavior Research Methods, 43, 602-615. 39: + 40: +They apply a "random" but invertible permutation before doing the 41: +circular convolution, to force it to be non-commutative. 42: + 43: +Compare their formulation to our original source.