Lines Matching full:aggregation
91 // This side channel implements aggregation via sum over a set of
100 // Dimensions of the aggregand. This is used by the secure aggregation
139 // quantization should target). The Secure Aggregation bitwidth (i.e.,
193 // A metric representing one stat with aggregation type sum.
196 // A metric representing a ratio between metrics with aggregation
205 // A metric representing one stat with aggregation type only sample.
239 // A metric representing one stat with aggregation type sum.
253 // A metric representing a ratio between metrics with aggregation type average.
271 // A metric representing one stat with aggregation type none.
277 // A metric representing one stat with aggregation type only sample.
372 // Minimum number of clients in aggregation.
373 // In secure aggregation mode this is used to configure the protocol instance
376 // Without secure aggregation server still respects this parameter,
429 // protocol, e.g. input checkpoints or aggregation protocols. This is a "tensors
430 // in, tensors out" interface. New aggregation methods can be added without
596 // use Secure Aggregation). If this field is not set, then the `ReportRequest`
599 // used to validate that the plan only uses Secure Aggregation.
610 // Describes which output tensors should be aggregated using an aggregation
625 // Describes how each output vector should be aggregated using an aggregation
637 // Aggregation, using the specified config options.
641 // support for simple aggregation without writing to an intermediate
648 // aggregation type (i.e. it cannot be used with TensorflowSpec output
661 // Parameters for the TFV1 Checkpoint Aggregation protocol.
664 // aggregation type (i.e. it cannot be used with TensorflowSpec output
876 // aggregation, and post-aggregation.
878 // The pre-broadcast and post-aggregation components are described with
888 // be used by the aggregation and post-aggregation logic.
890 // The aggregation logic represents the aggregation of client results at the
892 // Each ServerAggregationConfig message describes a single aggregation operation
898 // The post-aggregation logic is obtained by transforming the server_result TFF
907 // aggregation portion of this ServerPhaseV2 message alongside the pre- and
908 // post-aggregation logic from the original ServerPhase message. Ultimately,
929 // post-aggregation. This should be used with the TensorFlow graph defined
956 // for aggregation and the post-aggregation logic.
978 // Represents a single aggregation operation, combining one or more input
982 // The uri of the aggregation intrinsic (e.g. 'federated_sum').
985 // Describes an argument to the aggregation operation.
996 // List of arguments for the aggregation operation. The arguments can be
1004 // List of server-side outputs produced by the aggregation operation.
1007 // List of inner aggregation intrinsics. This can be used to delegate parts
1008 // of the aggregation logic (e.g. a groupby intrinsic may want to delegate
1016 // Represents a server phase which implements TF-based aggregation of multiple
1019 // There are two different modes of aggregation that are described
1020 // by the values in this message. The first is aggregation that is
1021 // coming from coordinated sets of clients. This includes aggregation
1022 // done via checkpoints from clients or aggregation done over a set
1023 // of clients by a process like secure aggregation. The results of
1024 // this first aggregation are saved to intermediate aggregation
1025 // checkpoints. The second aggregation then comes from taking
1028 // These two different modes of aggregation are done on different
1044 // L1 "Intermediate" Aggregation.
1046 // This is the initial aggregation that creates partial aggregates from client
1047 // results. L1 Aggregation may be run on many different instances.
1054 // Operation to run before the first aggregation happens.
1055 // For instance, clears the accumulators so that a new aggregation can begin.
1078 // Optionally restores variables where aggregation is done across
1082 // channels where aggregation is done by a separate process (such
1083 // as in secure aggregation), in which the side channel aggregated
1111 // End L1 "Intermediate" Aggregation.
1115 // L2 Aggregation and Coordinator.
1117 // This aggregates intermediate checkpoints from L1 Aggregation and performs
1119 // that does this aggregation.
1128 // This currently re-uses the `phase_init_op` from L1 aggregation above.
1154 // This is used for downstream, cross-round aggregation of metrics.
1179 // End L2 Aggregation.
1187 // This phase *does not include any aggregation.*
1280 // Optional. Server phase for TF-based aggregation; not provided for
1284 // Optional. Server phase for native aggregation; only provided for tasks
1306 // version >= 1 - plan supports multi-shard aggregation mode (L1/L2)
1346 // Represents the cross round aggregation portion for user defined measurements.
1348 // after a round of computation, to achieve aggregation beyond a round.