Sam Corbett-Davies

Algorithmic Decision Making and the Cost of Fairness

Algorithms are now regularly used to decide whether defendants awaiting trial are too dangerous to be released back into the community. In some cases, black defendants are substantially more likely than white defendants to be incorrectly classified as high risk. To mitigate such disparities, several techniques have recently been proposed to achieve algorithmic fairness. Here we reformulate algorithmic fairness as constrained optimization: the objective is to maximize public safety while satisfying formal fairness constraints designed to reduce racial disparities. We show that for several past definitions of fairness, the optimal algorithms that result require detaining defendants above race-specific risk thresholds. We further show that the optimal unconstrained algorithm requires applying a single, uniform threshold to all defendants. The unconstrained algorithm thus maximizes public safety while also satisfying one important understanding of equality: that all individuals are held to the same standard, irrespective of race. Because the optimal constrained and unconstrained algorithms generally differ, there is tension between improving public safety and satisfying prevailing notions of algorithmic fairness. By examining data from Broward County, Florida, we show that this trade-off can be large in practice. We focus on algorithms for pretrial release decisions, but the principles we discuss apply to other domains, and also to human decision makers carrying out structured decision rules.

Enriching object detection with 2D-3D registration and continuous viewpoint estimation

A large body of recent work on object detection has focused on exploiting 3D CAD model databases to improve detection performance. Many of these approaches work by aligning exact 3D models to images using templates generated from renderings of the 3D models at a set of discrete viewpoints. However, the training procedures for these approaches are computationally expensive and require gigabytes of memory and storage, while the viewpoint discretization hampers pose estimation performance. We propose an efficient method for synthesizing templates from 3D models that runs on the fly - that is, it quickly produces detectors for an arbitrary viewpoint of a 3D model without expensive dataset-dependent training or template storage. Given a 3D model and an arbitrary continuous detection viewpoint, our method synthesizes a discriminative template by extracting features from a rendered view of the object and decorrelating spatial dependences among the features. Our decorrelation procedure relies on a gradient-based algorithm that is more numerically stable than standard decomposition-based procedures, and we efficiently search for candidate detections by computing FFT-based template convolutions. Due to the speed of our template synthesis procedure, we are able to perform joint optimization of scale, translation, continuous rotation, and focal length using Metropolis-Hastings algorithm. We provide an efficient GPU implementation of our algorithm, and we validate its performance on 3D Object Classes and PASCAL3D+ datasets.

The Problem of Infra-Marginality in Outcome Tests for Discrimination

In the course of conducting traffic stops, officers have discretion to search motorists for drugs, weapons, and other contraband. There is concern that these search decisions are prone to racial bias, but it has proven difficult to rigorously assess claims of discrimination. Here we develop a new statistical method---the threshold test---to test for racial discrimination in motor vehicle searches. We use geographic variation in stop outcomes to infer the effective race-specific standards of evidence that officers apply when deciding whom to search, an approach we formalize with a hierarchical Bayesian latent variable model. This technique mitigates the problems of omitted variables and infra-marginality associated with benchmark and outcome tests for discrimination. On a dataset of 4.5 million police stops in North Carolina, we find that the standard for searching black and Hispanic drivers is considerably lower than the standard for searching white and Asian drivers, a pattern that holds consistently across the 100 largest police departments in the state.In the course of conducting traffic stops, officers have discretion to search motorists for drugs, weapons, and other contraband. There is concern that these search decisions are prone to racial bias, but it has proven difficult to rigorously assess claims of discrimination. Here we develop a new statistical method --- the threshold test --- to test for racial discrimination in motor vehicle searches. We use geographic variation in stop outcomes to infer the effective race-specific standards of evidence that officers apply when deciding whom to search, an approach we formalize with a hierarchical Bayesian latent variable model. This technique mitigates the problems of omitted variables and infra-marginality associated with benchmark and outcome tests for discrimination. On a dataset of 4.5 million police stops in North Carolina, we find that the standard for searching black and Hispanic drivers is considerably lower than the standard for searching white and Asian drivers, a pattern that holds consistently across the 100 largest police departments in the state.

An advanced interaction framework for augmented reality based exposure treatment

In this paper we present a novel interaction framework for augmented reality, and demonstrate its application in an interactive AR exposure treatment system for the fear of spiders. We use data from the Microsoft Kinect to track and model real world objects in the AR environment, enabling realistic interaction between them and virtual content. Objects are tracked in three dimensions using the Iterative Closest Point algorithm and a point cloud model of the objects is incrementally developed. The approximate motion and shape of each object in the scene serve as inputs to the AR application. Very few restrictions are placed on the types of objects that can be used. In particular, we do not require objects to be marked in a certain way in order to be recognized, facilitating natural interaction. To demonstrate our interaction framework we present an AR exposure treatment system where virtual spiders can walk up, around, or behind real objects and can be carried, prodded and occluded by the user. We also discuss improvements we are making to the interaction framework and its potential for use in other applications.

An interactive Augmented Reality system for exposure treatment

In this poster we describe an Augmented Reality (AR) system we are developing for exposure treatment. AR has great potential for phobia treatment because virtual fear stimuli can be shown in the real world and the client can see their own body and interact naturally with the stimuli. However, advanced natural interactivity has so far not been fully implemented in AR based exposure therapy systems. The novelty of our approach is based on better integrating the real environment and the user into the system, and in recognising natural user actions as system input. Using the Microsoft Kinect device, we create a model of the therapy environment and the users body. This information is used in conjunction with a physics simulation engine to create a virtual spider that reacts to the real environment in a realistic manner. The virtual spider can walk up, around, or behind real objects and can be carried, prodded and occluded by the user. We present the most recent prototype of the system and discuss the improvements we continue to make.

An expert system for automatically pruning vines

Vine pruning is an important part of vineyard management, and pruning is the most expensive task in the vineyard which has not yet been automated. Every year, most new canes must be removed from the vine, and the choice of canes to retain impacts vine yield. To automate the process of vine pruning, a vine pruning robot must make decisions on what canes to remove or to keep, based on a 3D topological model of the structure of the vine. In this paper we present an Artificial Intelligence (AI) system for making these decisions, developed and evaluated using simulated vines. A viticulture expert evaluated our approach by comparing it to pruning decisions made by a pruner with a skill level typical of human pruners. Our system successfully pruned 30% of vines better than the human and 89% at least as well. These results demonstrate that the vine pruning problem is solvable using current computing technologies, and that automating the pruning process has the potential to improve vine quality and yield.

Physically interactive tabletop augmented reality using the Kinect

In this paper we present a method for allowing arbitrary objects to interact physically in an augmented reality (AR) environment. A Microsoft Kinect is used to track objects in 6 degrees of freedom, enabling realistic interaction between them and virtual content in an tabletop AR context. We propose a point cloud based method for achieving such interaction. An adaptive per-pixel depth threshold is used to extract foreground objects, which are grouped using connected-component analysis. Objects are tracked with a variant of the Iterative Closest Point algorithm, which uses randomised projective correspondences. Our algorithm tracks objects moving at typical tabletop speeds with median drifts of 8.5% (rotational) and 4.8% (translational). The point cloud representation of foreground objects is improved as additional views of the object are visible to the Kinect. Physics-based AR interaction is achieved by fitting a collection of spheres to the point cloud model and passing them to the Bullet physics engine as a physics proxy of the object. Our method is demonstrated in an AR application where the user can interact with a virtual tennis ball, illustrating our proposed methods potential for physics-based AR interaction.

A Robot System for Pruning Grape Vines

This paper describes a robot system for the automatic pruning of grape vines. A mobile platform straddles the row of vines, and it images them with trinocular stereo cameras as it moves. A computer vision system builds a three-dimensional (3D) model of the vines, an artificial intelligence (AI) system decides which canes to prune, and a six degree-of-freedom robot arm makes the required cuts. The system is demonstrated cutting vines in the vineyard. The main contributions of this paper are the computer vision system that builds 3D vine models, and the test of the complete-integrated system. The vine models capture the structure of the plants so that the AI system can decide where to prune, and they are accurate enough that the robot arm can reach the required cuts. Vine models are reconstructed by matching features between images, triangulating feature matches to give a 3D model, then optimizing the model and the robots trajectory jointly (incremental bundle adjustment). Trajectories are estimated online at 0.25 m/s, and they have errors below 1% when modeling a 96 m row of 59 vines. Pruning each vine requires the robot arm to cut an average of 8.4 canes. A collision-free trajectory for the arm is planned in intervals of 1.5 s/vine with a rapidly exploring random tree motion planner. The total time to prune one vine is 2 min in field trials, which is similar to human pruners, and it could be greatly reduced with a faster arm. Trials also show that the long chain of interdependent components limits reliability. A commercially feasible pruning robot should stop and prune each vine in turn.

Interactive AR exposure therapy

In this demonstration we show an Augmented Reality (AR) system we are developing for exposure treatment. AR has great potential for phobia treatment because virtual fear stimuli can be shown in the real world and the client can see their own body and interact naturally with the stimuli. However, advanced natural interactivity has so far not been fully implemented in AR based exposure therapy systems. Our AR exposure treatment system has a better integration of the real environment and the user into the system, and recognizes natural user actions as system input. Using the Microsoft Kinect device, we create a model of the therapy environment and the users body. This information is used in conjunction with a physics simulation engine to create a virtual spider that reacts to the real environment in a realistic manner. The virtual spider can walk up, around, or behind real objects and can be carried, prodded and occluded by the user. We describe the system and present the iterative development of our framework including an improved gesture library for improved interactivity.