Riding with the Stars: Passenger Privacy in the NYC Taxicab Dataset

In my previous post, Differential Privacy: The Basics, I provided an introduction to differential privacy by exploring its definition and discussing its relevance in the broader context of public data release. In this post, I shall demonstrate how easily privacy can be breached and then counter this by showing how differential privacy can protect against this attack. I will also present a few other examples of differentially private queries.

The Data

There has been a lot of online comment recently about a dataset released by the New York City Taxi and Limousine Commission. It contains details about every taxi ride (yellow cabs) in New York in 2013, including the pickup and drop off times, locations, fare and tip amounts, as well as anonymized (hashed) versions of the taxi’s license and medallion numbers. It was obtained via a FOIL (Freedom of Information Law) request earlier this year and has been making waves in the hacker community ever since.

The release of this data in this unalloyed format raises several privacy concerns. The most well-documented of these deals with the hash function used to “anonymize” the license and medallion numbers. A bit of lateral thinking from one civic hacker and the data was completely de-anonymized. This data can now be used to calculate, for example, any driver’s annual income. More disquieting, though, in my opinion, is the privacy risk to passengers. With only a small amount of auxiliary knowledge, using this dataset an attacker could identify where an individual went, how much they paid, weekly habits, etc. I will demonstrate how easy this is to do in the following section.

Violating Privacy

Let’s consider some of the different ways in which this dataset can be exploited. If I knew an acquaintance or colleague had been in New York last year, I could combine known information about their whereabouts to try and track their movements for my own personal advantage. Maybe they filed a false expense report? How much did they tip? Did they go somewhere naughty? This can be extended to people I don’t know – a savvy paparazzo could track celebrities in this way, for example.

There are other ways to go about this too. Simply focusing the search on an embarrassing night spot, for example, opens the door to all kinds of information about its customers, such as name, address, marital status, etc. Don’t believe me? Keep reading…

Stalking celebrities

First things first. How might I track a person? Well, to zone in on a particular trip, I can use any combination of known characteristics that appear in the dataset, such as the pickup or drop-off coordinates or datetime, the medallion or license number, or even the fare amount from a receipt. Being the avid fanboy that I am (note: sarcasm), I thought it might be interesting to find out something new about some of the celebrities who had been seen in New York in 2013. In particular, where did they go to / come from, and how much did they tip?

In order to do this, I spent some of the most riveting hours of my professional career searching through images of “celebrities in taxis in Manhattan in 2013” to find enough information to identify the correct record in the database. I had some success – combining the below photos of Bradley Cooper and Jessica Alba with some information from celebrity gossip blogs allowed me to find their trips, which are shown in the accompanying maps.

Bradley Cooper (Click to Explore)

Jessica Alba (Click to Explore)

In Brad Cooper’s case, we now know that his cab took him to Greenwich Village, possibly to have dinner at Melibea, and that he paid $10.50, with no recorded tip. Ironically, he got in the cab to escape the photographers! We also know that Jessica Alba got into her taxi outside her hotel, the Trump SoHo, and somewhat surprisingly also did not add a tip to her $9 fare. Now while this information is relatively benign, particularly a year down the line, I have revealed information that was not previously in the public domain. Considering the speculative drivel that usually accompanies these photos (trust me, I know!), a celebrity journalist would be thrilled to learn this additional information.

A few innocent nights at the gentlemen’s club

But OK, perhaps you’re not convinced. After all, this dataset is (thankfully) not real-time. How about we leave the poor celebrities alone and consider something a little more provocative. Larry Flynt’s Hustler Club is in a fairly isolated location in Hell’s Kitchen, and no doubt experiences significant cab traffic in the early hours of the morning. I ran a query to pull out all pickups that occurred outside the club after midnight and before 6am, and mapped the drop-off coordinates to see if I could pinpoint individuals who frequented the establishment. The map below shows my results – the yellow points correspond to drop-offs that are closely clustered, implying a frequent customer.

Click to Explore

The potential consequences of this analysis cannot be overstated. Go ahead, zoom in. You will see that the GPS coordinates are terrifyingly precise. Using this freely-obtainable, easily-created map, one can find out where many of Hustler’s customers live, as there are only a handful of locations possible for each point. Add a little local knowledge, and, well, it’s not rocket science. “I was working late at the office” no longer cuts it: Big Brother is watching.

Even without suspicions or knowledge of the neighborhood, I was able to pinpoint certain individuals with high probability. Somewhat shockingly, just googling an address reveals all kinds of information about its inhabitants. Take the following example:

Examining one of the clusters in the map above revealed that only one of the 5 likely drop-off addresses was inhabited; a search for that address revealed its resident’s name. In addition, by examining other drop-offs at this address, I found that this gentleman also frequented such establishments as “Rick’s Cabaret” and “Flashdancers”. Using websites like Spokeo and Facebook, I was also able to find out his property value, ethnicity, relationship status, court records and even a profile picture!

Of course, I will not publish any of this information here, but this was by no means a unique situation. While the online availability of all this potentially private information is part of a wider discussion, it’s fair to say that this guy has a right to keep his nighttime activities a secret.

To reiterate: the findings in this section were not hard to uncover. Equipped with this dataset, and just a little auxiliary information about you, it would be quite trivial for someone to follow your movements, collecting data on your whereabouts and habits, while you remain blissfully unaware. A stalker could find out where you live and work. Your partner may spy on you. A thief could work out when you’re away from home, based on your habits. There are obvious mitigating factors here, such as the population density of Manhattan and the time delay of the data release, but the point still stands.

Applying Differential Privacy

So, we’re at a point now where we can agree this data should not have been released in its current form. But this data has been collected, and there is a lot of value in it – ask any urban planner. It would be a shame if it was withheld entirely.

Enter differential privacy. Sure, the data could be anonymized, but that obviously didn’t work out too well. More care could be taken, for example hashes could be salted as suggested in this forum. But that still doesn’t really protect against either of the attacks above. If you read my first post, you will know that only differential privacy guarantees the protection of all individuals in the data.

So how should we go about applying differential privacy? Remember that differential privacy works by adding noise to the query. Our three maps above are point queries: they filter the data rather than aggregating it. This means we have to be extra careful. We cannot simply run the query and add noise to the result (as we might do with an aggregation query) since this will release private information! To understand why, suppose we framed our question like this: “How many taxis picked up a passenger outside The Greenwich Hotel at 19:35 on July 8, 2013?” By running our original query, even after adding noise to the output coordinates, we have indirectly answered this question accurately. This constitutes a privacy breach, as information about specific individuals can be learned in this way.

Instead, we have to turn our query into an aggregation. If we place a grid over our map, and count the number of output coordinates that fall into each cell, we end up with a set of counts that are generally independent of each other. Differential privacy is now straightforward to apply, as we are working with aggregate quantities. The privatized versions of these three queries are displayed below. Please refer to the appendix for a more technical discussion of how this was done.

We can appreciate from these privatized maps the importance of ε, the privacy parameter, which I introduced in Differential Privacy: The Basics. When ε is low we can no longer accurately track either celebrity, nor learn how much they spent. In fact, it takes unreasonably high levels of ε to reliably locate them. We could opt for a finer grid, indeed we could all but replicate the point query with a fine enough grid, but differential privacy’s Laplace mechanism is robust enough to effectively ensure that the noise obfuscates any actual difference in the data. The privatization of the Hustler query is more interesting – since it is less targeted, the difference caused by the participation of one individual is less pronounced. As a result, there is more information in the privatized output – for example, the square over Wall Street still stands out, which echoes the actual evidence shown above.

Cooper, Privatized
(Click to Interact)

Alba, Privatized
(Click to Interact)

Hustler Customers, Privatized
(Click to Interact)

What about other queries? After all, did we just turn the data into nonsense? Clearly, our ability to learn what we sought from these queries was severely compromised. However, since they were so targeted, one could argue it is a good thing that the results are hard to interpret. The visualization below shows the results of other queries, and demonstrates that useful insights may still be extracted, particularly when aggregating over all records.

Click to Interact

Concluding Remarks

Now that you’re an expert in differential privacy, I urge you to consider: what should have been released by the NYC Taxi and Limousine Commission? How should the public be able to access it? Is this in line with Freedom of Information Law? If not, how should the law be changed to accommodate private data release?

It is only by considering these questions that we can hope to motivate a change in practices. With data privacy and security mishaps cropping up in the news almost daily, the topic has never before received so much attention. This post is yet another reminder of the fallibility of our systems and practices. As we know, differential privacy, with its strong guarantees – borne out of its sturdy fundamentals – offers a solution to many of these concerns.

As data scientists, it is in our best interests to encourage free data flow in our society, and lead the charge to ensure participants in these datasets are protected. The science is clearly there, and will continue to be developed in the coming years. The onus is now on industry and governments to pay attention and embrace differential privacy to protect their stakeholders and citizens.

Appendices

SQL queries used in this analysis

Tracking Bradley Cooper and Jessica Alba:

SELECT D.dropoff_latitude, D.dropoff_longitude, F.total_amount, F.tip_amount
FROM tripData AS D, tripFare AS F
WHERE D.hack_license = F.hack_license AND D.pickup_datetime = F.pickup_datetime
  AND pickup_datetime > "2013-07-08 19:33:00" AND pickup_datetime < "2013-07-08 19:37:00"
  AND pickup_latitude > 40.719 AND pickup_latitude < 40.7204
  AND pickup_longitude > -74.0106 AND pickup_longitude < -74.01;
SELECT D.dropoff_latitude, D.dropoff_longitude, F.total_amount, F.tip_amount
FROM tripData AS D, tripFare AS F
WHERE D.hack_license = F.hack_license AND D.pickup_datetime = F.pickup_datetime
  AND dropoff_datetime > "2013-09-07 12:19:00" AND dropoff_datetime < "2013-09-07 12:25:00"
  AND dropoff_latitude > 40.727 AND dropoff_latitude < 40.728
  AND dropoff_longitude > -73.994 AND dropoff_longitude < -73.993;

Identifying Hustler customers:

SELECT dropoff_latitude, dropoff_longitude
FROM tripData
WHERE pickup_latitude > 40.767249 AND pickup_latitude < 40.768
  AND pickup_longitude > -73.996782 AND pickup_longitude < -73.995538
  AND HOUR(pickup_datetime) < 6
  AND trip_distance > 5;
Privacy calculations

I have chosen to use the Laplace mechanism to privatize the above 3 queries, as it is relatively easy to apply and explain. However, in general, the Laplace mechanism is not appropriate for geospatial data. For example, it does not consider topographical features – inspections of the privatized maps show positive counts in the middle of the East River! Rather, there are more complex methods for adding noise to spatial data – Graham Cormode’s paper, Differentially Private Spatial Decompositions, offers a more thorough mechanism, while this blog covers the topic more generally. However, I will proceed with Laplace, in the hope that, by moving away from discussions of complicated mechanisms, the core tenets of differential privacy may be grasped more easily.

So how should we go about privatizing the above queries using the Laplace mechanism? As mentioned above, these are point queries. We should not apply noise directly to the output, because our sensitivity is essentially undefined. Why? Let’s consider the identity query, Q(D) = D. This is a point query – there is no aggregation. What happens if we remove a record from the dataset? We have Q(D’) = D’. Fine. But now look at what happens to our sensitivity. Recall the formula:

\Delta \mathit{f}=\underset{D,D'}{max}||\mathit{Q(D)}-\mathit{Q(D')}||_{1}

This formula is only defined when Q() returns a vector of fixed length (independent of D), such as an average, or a sum. With Q(D)=D, this is of course not the case, and so we cannot calculate the sensitivity. Intuitively, this result is true for other point queries. So we need to find another way to privatize point queries.

We know that privatizing aggregation queries is a lot more straightforward. Is there a way to turn our point query into an aggregation query? What if we bucketed the data into similar groups and then counted how many records returned by the query fall into each group? If we define our groups tightly enough, we would fully replicate the point query. We end up with a series of independent counts, which is easily privatized – the change from removing one record, any record, is at most 1. So this is our sensitivity. Adding Laplace noise with a sensitivity of 1 to the count in each bucket guarantees differential privacy.

Of course, if we have too many buckets, we will have lots of 0s and 1s. Since adding Laplace noise to this essentially hides this difference (when ε is at a reasonable level of course), we may struggle to learn anything from this data. By allowing each bucket to incorporate a greater range of values, we can obtain more meaningful privatized counts, albeit by sacrificing a degree of accuracy. The optimal way to group the data therefore depends on which features of the output are of interest.

This method is not confined to one-dimensional responses. As can be seen in the privatized maps above, the histogram is 2-dimensional, reflecting the latitude / longitude output from each query. In fact, any mapping of n-dimensional data into independent buckets is sufficient to ensure differential privacy. For the celebrity queries, I did treat the latitude/longitude coordinates and the fare/tip amounts as two separate histogram queries, as these pairs are broadly uncorrelated with each other.

For additional detail on the privacy calculations used throughout this post, the reader is encouraged to examine the code behind the various visualizations. While it is clear that the Laplace mechanism is applied slightly differently each time, it is not hard to imagine how this might be automated so differential privacy could be applied more generally.

Differential Privacy: The Basics

Hi! I’m Anthony Tockar. I am a masters student at Northwestern University and have been working with the Science team for the summer. This is the first of two posts I will contribute on the topic of differential privacy.

Anyone who has ever googled themselves, updated their Facebook settings or even received a call from a telemarketer knows about the importance of privacy. In a world where the data being collected on us is growing exponentially, attacks on our privacy are becoming commonplace. A quick search will reveal all kinds of horror stories, such as the identification of individuals’ AOL search queries, people’s movie ratings on Netflix, and the medical records of a former Massachusetts governor. In my next post, I will demonstrate how easy it is to track individuals in New York using data from the 2013 NYC Taxi data release. Clearly, common practices need to change when it comes to protecting the individual. Put another way, is it fair to citizens and consumers to use their data without even being able to guarantee their privacy?

Data custodians are acutely aware of this issue. Here at Neustar, for example, our datasets span many domains, and can often contain sensitive information, such as PII or “harmless” DNS lookups. Our customers, of course, trust us to keep this information private. Therefore it is not surprising that we employ a series of privacy experts to maintain the company’s contracts and policies with respect to data. As you might expect, it is also of interest to us in the data science team to know what rigor has been applied to this problem.

So – how is data privatized? It turns out that there are various methods that are commonly found in practice, from simple fixes like removing columns containing PII, to more advanced treatments such as k-anonymity and l-diversity. ALL of these methods have been shown to be vulnerable to attack in one way or another. However, in the last few years a new method has emerged that has survived close scrutiny: differential privacy. Unlike other methods, differential privacy operates off a solid mathematical foundation, making it possible to provide strong theoretical guarantees on the privacy and utility of released data.

The story of differential privacy, however, is not all good news. Despite showing massive potential, it has been used sparingly in practice, and remains largely confined to theoretical settings. In this post I hope to introduce the topic to a wider audience, and have included some interactive visualizations to help readers build their intuition. I will also discuss the relevance of differential privacy in more detail, and in doing so explore whether it really is the solution to the privacy puzzle.

Definition of Differential Privacy

This simple example should help illuminate the concept:

Suppose you have access to a database that allows you to compute the total income of all residents in a certain area. If you knew that Mr. White was going to move to another area, simply querying this database before and after his move would allow you to deduce his income.

What could one do to stop this? Perhaps make sure that every query returns an approximation of the total income? This could accomplish the goal of allowing the dataset to yield accurate information while protecting Mr. White. Formally, differential privacy is defined as follows:

A randomized function K gives ε-differential privacy if for all data sets D and D′ differing on at most one row, and all S ⊆ Range(K), Pr[K(D)\in S]\leq exp(\varepsilon)\times Pr[K(D')\in S]

This can be translated as meaning that the risk to one’s privacy should not substantially (as bounded by ε) increase as a result of participating in a statistical database. Thus an attacker should not be able to learn any information about any participant that they could not learn if the participant had opted out of the database. One could then state with some confidence that there is a low risk of any individual’s privacy being compromised as a result of their participation in the database.

Sounds great! So, how is it applied? Let’s go back to our example. If we can add some noise to the result of a query on our dataset, we should be able to ensure that the formula above holds. The function K() is our mechanism for adding this noise. Differential privacy-preserving mechanism design is a topic on its own. Here it suffices to say that there are different mechanisms available depending on the use case: for example, the Laplace mechanism, which I will introduce in the next paragraph, is unsuitable for categorical data – the exponential mechanism is more appropriate, though a strong case can be made for other methods depending on the outcome.

The Laplace mechanism involves adding random noise that conforms to the Laplace statistical distribution. That’s it! The obvious follow-up question is: well, how much noise should be added – i.e. how should we define our Laplace random variable? The 0-centered Laplace distribution has only one parameter (its scale), and this is directly proportional to its standard deviation, or noisiness. So how should we set the scale? Naturally it should have some dependence on the privacy parameter, ε. It should also depend on the nature of the query itself, and more specifically, the risk to the most different individual of having their private information teased out of the data. This can be defined mathematically, and is known as the sensitivity of the query:

\Delta \mathit{f}=\underset{D,D'}{max}||\mathit{f(D)}-\mathit{f(D')}||_{1}

Simply stated, it is the maximum difference in the values that the query f may take on a pair of databases that differ in only one row. There is a neat proof that shows that by adding a random Laplace(\Delta \mathit{f} / \varepsilon) variable to a query, ε-differential privacy is guaranteed.

The more astute among us may be thinking: “Wait! If you just add a symmetrical, 0-centred distribution to the data, I can just run the query multiple times and take the average!” You’d be right if not for the property known as composition. Now while the definition differs depending on the technique used to query the data, in the simple, non-adaptive case, the composition property is as follows:

For a dataset queried q times, with each query having privacy parameter εi, the total privacy budget of the dataset is given by \varepsilon _{total} =\sum_{i=1}^{q}\varepsilon _{i}.

So when we think about ε, we should really be thinking about a privacy budget rather than purely the statistical upper bound of a query. Sure, each query yields \varepsilon _{i}, but it is the total ε we should be concerned about as it reflects the maximum privacy release allowable for the total query session. As each query answer leaks privacy, once the budget is exceeded the user will not be able to make any further queries. It is this feature that allows differential privacy to work in practice.

How might this look?

Since we now have a basic grasp of the concept, let’s take it for a spin! In our Mr. White example above, let’s assume the total income in his original neighborhood is $50 million. After he leaves, this figure drops to $49 million. Therefore, one can infer that his true income is $1 million. To keep his income private, we have to ensure the query response is noisy enough to ‘hide’ this information. In fact, to ensure we privatize income for all people in our dataset we need to make sure the richest person is protected as well. As it turns out, Mr. White was the richest person in his neighborhood, so the sensitivity is $1 million *. Play around with the tool below to see what this looks like in practice – I have left ε variable to emphasize its effect.

Click to Interact

It is plain to see that Mr. White is protected – even at lenient levels of privacy, it is all but impossible to deduce his true income by running the query once. The histogram shows what would happen if we were given free rein to run multiple queries – eventually, we would reach a point where we could easily find his true income. However, we would be violating our \varepsilon _{total} requirement as mentioned above.

Now that we’ve seen what would happen with a query on a single value, let’s look at what happens when we privatize a whole distribution of values. The tool below provides the density plot for several different distributions, both real and randomly-generated, and again the user is free to experiment with different privacy parameters. Hopefully through interacting with this widget, the reader can develop a more intuitive understanding of the effects of differential privacy.

Click to Interact

There is a key difference in how this query is privatized. As it is a density plot, we are only interested in how many observations occur at each value (similar to a histogram). Since this is a series of counts, our sensitivity is 1 (the maximum amount an individual can contribute). It follows that we can add Laplace random variables to the count at each value to guarantee privacy. Click on the graph for a more in-depth description.

Note that this definition of sensitivity is dependent on our definition of the dataset – if we were considering every neighborhood in the US for example our sensitivity would be a lot higher – but we could adjust for this by setting a more relaxed privacy parameter given the difficulty of prizing an individual’s private data from a dataset of this breadth.

Relevance of Differential Privacy

Why should we care? At the very least differential privacy provides an interesting and rigorous framework around publishing data. So far, it is the only approach that has both theoretical bounds on the privacy of users in the dataset and still enables scientists to mine useful insights from it. Other solutions, such as anonymization (hashing), have time and again proven insufficient. Just look at the recent NYC taxi data that was de-anonymized within days of its release! Even with perfect anonymization, it has been posited that 63% of people in the US can be uniquely identified by just their birth date, zip code and gender.*

Unfortunately these attacks tend to stifle the release of data and information sharing. Differential privacy provides some hope. As we have learned, it is inherently flexible, which means it can easily be adapted to environments with differing privacy requirements. This flexibility does come at a cost: as we have seen, having very tiny privacy budgets (ε) can make some queries all but useless. However, as more people understand the concepts and more products get built on top of this paradigm we expect to see more sharing of data into the public domain without privacy concerns.

The more philosophical question is how private is private enough? Clearly, there is some tunability between how useful a differentially private query is and how ‘private’ it is. The aforementioned tradeoff between utility and privacy is unfortunately ‘left to the reader’. The literature does provide some rules of thumb for setting ε, with suggestions like 0.01, or ln2, etc. – however these have scant theoretical support. Perhaps most importantly, there are few, if any, precedents. At the end of the day it is the data curator’s job (or his lawyer) to decide on ‘private enough’. The lack of a clear framework to relate ε to privacy levels coupled with the difficulty of explaining it to the layperson has meant that differential privacy has largely remained confined to academia. However, as more people learn about it and more tools begin to emerge (PINQAiravat) this is starting to change. Clearly in this world of massive data sets and smart data scientists and hackers, data privacy needs to keep pace. We are very hopeful that these techniques are the next step.

* Aside: This is particularly interesting considering the birthday paradox, which states that in a room of 23 people, there is a >50% chance at least two of them share a birthday. However, by dissecting further by zip code and gender (and of course birth year), it appears that this no longer holds to the same degree.

Conclusion

Of most interest to us at Neustar is the practical implementation of differential privacy. For example, how can it be applied to the world of Internet advertising? What about sketching? Well, there has actually been some proprietary work in these areas. Our friend Muthu has used sketches to develop pan-private algorithms for dynamic data, which guarantee privacy and security for certain queries. In addition, this paper suggests differentially private measures for preventing attacks that utilize personalized advertising campaigns. Works such as these help pave the way for a new dawn in the way we think about data.

It is worth pointing out again that this post is not supposed to be a complete discussion of differential privacy and its nuances. Differential privacy comes in many different forms and iterations which have not been covered, and it does have several limitations. There have been reams written about it, and the interested reader is encouraged to seek out further information. The concept of differential privacy is still young, and there is still a lot to be fleshed out. What is clear is that it holds much potential.

In my next post, I will make it abundantly clear why we should be taking privacy seriously, and follow this up with a demonstration of how to privatize a real dataset in a way that protects its participants while retaining its ability to bestow useful insights.