Skip to main content
Search

École Polytechnique Fédérale de Lausanne

2 related articles

Subscribe to updates
  1. arXiv

    Output parity alone misses how predictions are built: adding an attribution-disparity penalty improves all six medical-imaging fairness trade-offs

    Fairness in medical image classification is usually enforced only on output-level group gaps, yet with a linear classification head the output score gap equals the signed sum of feature-attribution differences, so opposing contributions can cancel and make a model look fair. Adding the same attribution-disparity penalty to three equalized-odds surrogates produced positive paired hypervolume gains in all six comparisons on HAM10000 gender fairness and Harvard-GF race fairness, with relative gains of 1.49% to 10.57%. Under a predefined utility constraint, the augmented model families also cut equalized-odds difference by 9.0% to 38.7% while attribution disparity and the cancellation ratio fell.

  2. Sensor networks stop guessing only locally: nodes first find same-state neighbors, then recover the whole network

    In heterogeneous sensing networks, agents previously could infer only their own state or rely on a central processor to aggregate measurements. A decentralized method called cluster-aware social learning now lets each agent use local observations and neighbor exchanges to identify connected groups sharing the same hidden state, maintain one belief per group, and map those beliefs back to individual agents, recovering the network-wide state configuration while exchanging only cluster-level information. Synthetic Gaussian experiments and real gas-sensor data both show beliefs concentrating on each cluster's true state once the cluster structure is frozen.

You’re all caught up