Adam Jermyn

Quan­tiliz­ers and Gen­er­a­tive Models

Grouped Loss may dis­fa­vor dis­con­tin­u­ous capabilities

La­tent Ad­ver­sar­ial Training

Train­ing Trace Pri­ors and Speed Priors

Con­di­tion­ing Gen­er­a­tive Models

#SAT with Ten­sor Networks

Multi­gate Priors

Train­ing Trace Priors

ELK Pro­posal—Make the Re­porter care about the Pre­dic­tor’s beliefs

Give the model a model-builder

Give the AI safe tools