This project is mirrored from https://github.com/tensorflow/federated. Pull mirroring failed .
Last successful update .
  1. 24 8月, 2021 1 次提交
  2. 21 8月, 2021 3 次提交
  3. 20 8月, 2021 2 次提交
  4. 19 8月, 2021 3 次提交
  5. 18 8月, 2021 1 次提交
  6. 17 8月, 2021 1 次提交
  7. 14 8月, 2021 3 次提交
  8. 13 8月, 2021 2 次提交
  9. 12 8月, 2021 5 次提交
    • A. Unique TensorFlower's avatar
      Refactors the aggregator and tff computation for hierarchical histogram. · 31d16fa2
      A. Unique TensorFlower 创作于
      Changes:
      (1) Replaces the old dp_query `CentralTreeSumQuery` with the updated one `TreeRangeSumQuery`. The updated query has several advantages:
          1) It easily supports different DP mechanisms (e.g. GaussianSumQuery,
          DistributedDiscreteGaussianSumQuery) by taking them as inner queries;
          2) `TreeRangeSumQuery` does not hard-code any clipping inside so it is
          composable with various clipping factories. To avoid DP error, it does norm
         checking inside to make sure appropriate clipping happens outside.
      (2) Removes the type conversion from int32 to float32 before return in `discretized_histogram_counts`. Now the function outputs int32 tensors. This change is due to the type requirement of `HistogramClippingSumFactory`.
      (3) Updates tff's dependency on tensorflow privacy to 0.7.0.
      PiperOrigin-RevId: 390277449
      31d16fa2
    • Jakub Konecny's avatar
      Refactors optimizer_utils.py to remove one invocation of `model_fn` (from 4 to 3). · 39d795ba
      Jakub Konecny 创作于
      PiperOrigin-RevId: 390236191
      39d795ba
    • A. Unique TensorFlower's avatar
      Removes the non-negative enforcement in `enforce_consistency` because... · 54c4fc7a
      A. Unique TensorFlower 创作于
      Removes the non-negative enforcement in `enforce_consistency` because sometimes it introduces high bias.
      
      PiperOrigin-RevId: 390198511
      54c4fc7a
    • Keith Rush's avatar
      Returns second argument to binary operator in... · c84564db
      Keith Rush 创作于
      Returns second argument to binary operator in create_binary_operator_with_upcast uncoerced in the case of type assignability to the first arg.
      
      This change represents a slight relaxation in (static) semantics for create_binary_operator_with_upcast. Previously a binary operator with signature like:
      ```
      <float32[?],float32[1]>
      ```
      could not be constructed, though TensorFlow's shape assignability allows for a graph like this. This change allows for signatures such as the above to be represented in TF computation factory.
      
      We do not, however, open the door to signatures like:
      ```
      <float32[?],float32[1, 1]>
      ```
      (though TF will generate graphs which represent it), since such a signature does not pass TFF's notion of assignability.
      
      PiperOrigin-RevId: 390188849
      c84564db
    • Keith Rush's avatar
      Adds failing test for unknown shapes in TF computation factory binary operator with upcast. · bbded41f
      Keith Rush 创作于
      PiperOrigin-RevId: 390176777
      bbded41f
  10. 11 8月, 2021 3 次提交
  11. 10 8月, 2021 1 次提交
  12. 07 8月, 2021 2 次提交
  13. 05 8月, 2021 2 次提交
  14. 04 8月, 2021 2 次提交
  15. 03 8月, 2021 3 次提交
  16. 02 8月, 2021 1 次提交
    • Taylor Cramer's avatar
      Clean up and clarify `select_output_from_lambda` · f3ffe041
      Taylor Cramer 创作于
      The `select_output_from_lambda` function previously had
      surprisingly different semantics on outer-tuples from
      nested tuples: outermost tuples or lists indicated a
      structure of results, while inner tuples or lists
      indicated paths to elements. No distinction was made
      between tuples and lists.
      
      This CL changes to use tuples for multi-index
      selection paths and lists for selecting multiple
      different values. This simplifies some call sites.
      
      Includes test cleanup to remove outdated references
      to "Federated" and to test the new selection abilities.
      
      PiperOrigin-RevId: 388160914
      f3ffe041
  17. 31 7月, 2021 3 次提交
  18. 30 7月, 2021 2 次提交