Differentially personal heatmaps – Google AI Weblog

Just lately, differential privacy (DP) has emerged as a mathematically sturdy notion of person privateness for knowledge aggregation and machine studying (ML), with sensible deployments together with the 2022 US Census and in trade. Over the previous few years, we’ve open-sourced libraries for privacy-preserving analytics and ML and have been always enhancing their capabilities. In the meantime, new algorithms have been developed by the analysis neighborhood for a number of analytic duties involving personal aggregation of knowledge.

One such essential knowledge aggregation technique is the heatmap. Heatmaps are common for visualizing aggregated knowledge in two or extra dimensions. They’re broadly utilized in many fields together with laptop imaginative and prescient, picture processing, spatial knowledge evaluation, bioinformatics, and extra. Defending the privateness of person knowledge is vital for a lot of functions of heatmaps. For instance, heatmaps for gene microdata are primarily based on personal knowledge from people. Equally, a heatmap of common areas in a geographic space are primarily based on person location check-ins that have to be stored personal.

Motivated by such functions, in “Differentially Private Heatmaps” (offered at AAAI 2023), we describe an environment friendly DP algorithm for computing heatmaps with provable ensures and consider it empirically. On the core of our DP algorithm for heatmaps is an answer to the essential drawback of the right way to privately mixture sparse enter vectors (i.e., enter vectors with a small variety of non-zero coordinates) with a small error as measured by the Earth Mover’s Distance (EMD). Utilizing a hierarchical partitioning process, our algorithm views every enter vector, in addition to the output heatmap, as a likelihood distribution over various objects equal to the dimension of the info. For the issue of sparse aggregation underneath EMD, we give an environment friendly algorithm with error asymptotically near the very best.

Algorithm description

Our algorithm works by privatizing the aggregated distribution (obtained by averaging over all person inputs), which is adequate for computing a ultimate heatmap that’s personal attributable to the post-processing property of DP. This property ensures that any transformation of the output of a DP algorithm stays differentially personal. Our foremost contribution is a brand new privatization algorithm for the aggregated distribution, which we’ll describe subsequent.

The EMD measure, which is a distance-like measure of dissimilarity between two likelihood distributions initially proposed for laptop imaginative and prescient duties, is well-suited for heatmaps because it takes the underlying metric house into consideration and considers “neighboring” bins. EMD is utilized in quite a lot of functions together with deep studying, spatial evaluation, human mobility, picture retrieval, face recognition, visible monitoring, form matching, and extra.

To attain DP, we have to add noise to the aggregated distribution. We’d additionally wish to protect statistics at totally different scales of the grid to attenuate the EMD error. So, we create a hierarchical partitioning of the grid, add noise at every stage, after which recombine into the ultimate DP aggregated distribution. Particularly, the algorithm has the next steps:

  1. Quadtree development: Our hierarchical partitioning process first divides the grid into 4 cells, then divides every cell into 4 subcells; it recursively continues this course of till every cell is a single pixel. This process creates a quadtree over the subcells the place the basis represents your entire grid and every leaf represents a pixel. The algorithm then calculates the full likelihood mass for every tree node (obtained by including up the aggregated distribution’s chances of all leaves within the subtree rooted at this node). This step is illustrated under.
    In step one, we take the (non-private) aggregated distribution (prime left) and repeatedly divide it to create a quadtree. Then, we compute the full likelihood mass is every cell (backside).
  2. Noise addition: To every tree node’s mass we then add Laplace noise calibrated to the use case.
  3. Truncation: To assist cut back the ultimate quantity of noise in our DP aggregated distribution, the algorithm traverses the tree ranging from the basis and, at every stage, it discards all however the prime w nodes with highest (noisy) lots along with their descendants.
  4. Reconstruction: Lastly, the algorithm solves a linear program to get well the aggregated distribution. This linear program is impressed by the sparse recovery literature the place the noisy lots are considered as (noisy) measurements of the info.
In step 2, noise is added to every cell’s likelihood mass. Then in step 3, solely top-w cells are stored (inexperienced) whereas the remaining cells are truncated (purple). Lastly, within the final step, we write a linear program on these prime cells to reconstruct the aggregation distribution, which is now differentially personal.

Experimental outcomes

We consider the efficiency of our algorithm in two totally different domains: real-world location check-in knowledge and picture saliency knowledge. We contemplate as a baseline the ever present Laplace mechanism, the place we add Laplace noise to every cell, zero out any unfavourable cells, and produce the heatmap from this noisy mixture. We additionally contemplate a “thresholding” variant of this baseline that’s extra suited to sparse knowledge: solely preserve prime t% of the cell values (primarily based on the likelihood mass in every cell) after noising whereas zeroing out the remaining. To judge the standard of an output heatmap in comparison with the true heatmap, we use Pearson coefficient, KL-divergence, and EMD. Observe that when the heatmaps are extra related, the primary metric will increase however the latter two lower.

The areas dataset is obtained by combining two datasets, Gowalla and Brightkite, each of which comprise check-ins by customers of location-based social networks. We pre-processed this dataset to think about solely check-ins within the continental US leading to a ultimate dataset consisting of ~500,000 check-ins by ~20,000 customers. Contemplating the highest cells (from an preliminary partitioning of your entire house right into a 300 x 300 grid) which have check-ins from not less than 200 distinctive customers, we partition every such cell into subgrids with a decision of ∆ × ∆ and assign every check-in to one in all these subgrids.

Within the first set of experiments, we repair ∆ = 256. We take a look at the efficiency of our algorithm for various values of ε (the privacy parameter, the place smaller ε means stronger DP ensures), starting from 0.1 to 10, by working our algorithms along with the baseline and its variants on all cells, randomly sampling a set of 200 customers in every trial, after which computing the space metrics between the true heatmap and the DP heatmap. The common of those metrics is offered under. Our algorithm (the purple line) performs higher than all variations of the baseline throughout all metrics, with enhancements which can be particularly important when ε shouldn’t be too massive or small (i.e., 0.2 ≤ ε ≤ 5).

Metrics averaged over 60 runs when various ε for the placement dataset. Shaded areas point out 95% confidence interval.

Subsequent, we examine the impact of various the quantity n of customers. By fixing a single cell (with > 500 customers) and ε, we range n from 50 to 500 customers. As predicted by idea, our algorithms and the baseline carry out higher as n will increase. Nonetheless, the habits of the thresholding variants of the baseline are much less predictable.

We additionally run one other experiment the place we repair a single cell and ε, and range the decision ∆ from 64 to 256. In settlement with idea, our algorithm’s efficiency stays practically fixed for your entire vary of ∆. Nonetheless, the baseline suffers throughout all metrics as ∆ will increase whereas the thresholding variants sometimes enhance as ∆ will increase.

Impact of the variety of customers and grid decision on EMD.

We additionally experiment on the Salicon image saliency dataset (SALICON). This dataset is a group of saliency annotations on the Microsoft Common Objects in Context picture database. We downsized the pictures to a hard and fast decision of 320 × 240 and every [user, image] pair consists of a sequence of coordinates within the picture the place the person appeared. We repeat the experiments described beforehand on 38 randomly sampled photographs (with ≥ 50 customers every) from SALICON. As we are able to see from the examples under, the heatmap obtained by our algorithm may be very near the bottom reality.

Instance visualization of various algorithms for 2 totally different pure photographs from SALICON for ε = 10 and n = 50 customers. The algorithms from left to proper are: authentic heatmap (no privateness), baseline, and ours.

Extra experimental outcomes, together with these on different datasets, metrics, privateness parameters and DP fashions, will be discovered within the paper.


We offered a privatization algorithm for sparse distribution aggregation underneath the EMD metric, which in flip yields an algorithm for producing privacy-preserving heatmaps. Our algorithm extends naturally to distributed fashions that may implement the Laplace mechanism, together with the safe aggregation mannequin and the shuffle model. This doesn’t apply to the extra stringent local DP model, and it stays an attention-grabbing open query to plot sensible native DP heatmap/EMD aggregation algorithms for “reasonable” variety of customers and privateness parameters.


This work was executed collectively with Junfeng He, Kai Kohlhoff, Ravi Kumar, Pasin Manurangsi, and Vidhya Navalpakkam.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button