Gradient dataset aware configuration for over-the-air (OTA) model aggregation in federated learning
A method performed by a user equipment (UE) generates local gradients for a federated learning task. The method calculates a gradient Sum-Power level based on the local gradients. The method receives a mapping between gradient Sum-Power levels and scaling factors for channel inversion coefficients to process a data block into an unencoded uplink signal. The method also determines a channel inversion coefficient based on a scaling factor obtained from the mapping and the calculated gradient Sum-Power level. The method applies analog modulation and the channel inversion coefficient to the data block to form the unencoded uplink signal. The method further transmits, to a network, the unencoded uplink signal on shared uplink resources for an over-the-air computation of global gradients for the federated learning task.
1. A method of wireless communication performed by a user equipment (UE), comprising:
generating local gradients for a federated learning task;
calculating a gradient Sum-Power level based on the local gradients;
receiving a mapping between gradient Sum-Power levels and scaling factors for channel inversion coefficients to process a data block into an unencoded uplink signal;
determining a channel inversion coefficient based on a scaling factor obtained from the mapping and the calculated gradient Sum-Power level;
applying analog modulation and the channel inversion coefficient to the data block to form the unencoded uplink signal; and
transmitting, to a network, the unencoded uplink signal on shared uplink resources for an over-the-air computation of global gradients for the federated learning task.
2. The method of claim 1 , in which the mapping is associated with the federated learning task.
3. The method of claim 2 , in which the mapping comprises a range of Sum-Power levels assigned to each of the scaling factors.
4. The method of claim 3 , in which the federated learning task comprises a training task with a number of gradients, the number of gradients exceeding a threshold causing an increase of the range of Sum-Power levels assigned to each of the scaling factors.
5. The method of claim 1 , in which the mapping is defined in a telecommunications standard.
6. The method of claim 1 further comprising:
receiving a configuration for reporting the calculated gradient Sum-Power associated with a round of gradient over-the-air aggregation; and
reporting the calculated gradient Sum-Power based on the configuration.
7. The method of claim 6 , in which receiving the configuration occurs via a UE-specific downlink channel.
8. The method of claim 6 , in which receiving the configuration occurs via a group-common downlink channel.
9. The method of claim 6 , in which the configuration indicates a quantization level for reporting the calculated gradient Sum-Power.
10. The method of claim 6 , in which the configuration comprises an allocation of uplink resources for the reporting.
11. The method of claim 6 , in which the configuration corresponds to a plurality of rounds of the gradient over-the-air aggregation.
12. The method of claim 11 , in which the plurality of rounds comprise the overall federated learning task.
13. The method of claim 6 , in which the reporting occurs via a UE-specific uplink channel.
14. The method of claim 1 , in which receiving the mapping occurs via a group-common downlink channel associated with a selected gradient Sum-Power.
15. The method of claim 14 , in which the scaling factor corresponds to a reference signal received power (RSRP) value associated with the channel inversion coefficient.
16. The method of claim 15 , further comprising receiving the scaling factor along with a message triggering a reference signal associated with the RSRP value.
17. An apparatus for wireless communications performed by a user equipment (UE), comprising:
a processor,
memory coupled with the processor; and
instructions stored in the memory and operable, when executed by the processor, to cause the apparatus:
to generate local gradients for a federated learning task;
to calculate a gradient Sum-Power level based on the local gradients;
to receive a mapping between gradient Sum-Power levels and scaling factors for channel inversion coefficients to process a data block into an unencoded uplink signal;
to determine a channel inversion coefficient based on a scaling factor obtained from the mapping and the calculated gradient Sum-Power level;
to apply analog modulation and the channel inversion coefficient to the data block to form the unencoded uplink signal; and
to transmit, to a network, the unencoded uplink signal on shared uplink resources for an over-the-air computation of global gradients for the federated learning task.
18. The apparatus of claim 17 , in which the mapping is associated with the federated learning task.
19. The apparatus of claim 18 , in which the mapping comprises a range of Sum-Power levels assigned to each of the scaling factors.
20. The apparatus of claim 19 , in which the federated learning task comprises a training task with a number of gradients, the number of gradients exceeding a threshold causing an increase of the range of Sum-Power levels assigned to each of the scaling factors.
21. The apparatus of claim 17 , in which the mapping is defined in a telecommunications standard.
22. The apparatus of claim 17 , in which the processor causes the apparatus:
to receive a configuration for reporting the calculated gradient Sum-Power associated with a round of gradient over-the-air aggregation; and
to report the calculated gradient Sum-Power based on the configuration.
23. The apparatus of claim 22 , in which the processor causes the apparatus to receive the configuration via a UE-specific downlink channel.
24. The apparatus of claim 22 , in which the processor causes the apparatus to receive the configuration via a group-common downlink channel.
25. The apparatus of claim 22 , in which the configuration indicates a quantization level for reporting the calculated gradient Sum-Power.
26. The apparatus of claim 22 , in which the configuration comprises an allocation of uplink resources for the reporting.
27. The apparatus of claim 22 , in which the configuration corresponds to a plurality of rounds of the gradient over-the-air aggregation.
28. The apparatus of claim 27 , in which the plurality of rounds comprise the overall federated learning task.
29. The apparatus of claim 22 , in which the processor causes the apparatus to report via a UE-specific uplink channel.
30. The apparatus of claim 17 , in which the processor causes the apparatus to receive the mapping via a group-common downlink channel associated with a selected gradient Sum-Power.