-
Designing a Dashboard for Transparency and Control of Conversational AI
Authors:
Yida Chen,
Aoyu Wu,
Trevor DePodesta,
Catherine Yeh,
Kenneth Li,
Nicholas Castillo Marin,
Oam Patel,
Jan Riecke,
Shivam Raval,
Olivia Seow,
Martin Wattenberg,
Fernanda Viégas
Abstract:
Conversational LLMs function as black box systems, leaving users guessing about why they see the output they do. This lack of transparency is potentially problematic, especially given concerns around bias and truthfulness. To address this issue, we present an end-to-end prototype-connecting interpretability techniques with user experience design-that seeks to make chatbots more transparent. We beg…
▽ More
Conversational LLMs function as black box systems, leaving users guessing about why they see the output they do. This lack of transparency is potentially problematic, especially given concerns around bias and truthfulness. To address this issue, we present an end-to-end prototype-connecting interpretability techniques with user experience design-that seeks to make chatbots more transparent. We begin by showing evidence that a prominent open-source LLM has a "user model": examining the internal state of the system, we can extract data related to a user's age, gender, educational level, and socioeconomic status. Next, we describe the design of a dashboard that accompanies the chatbot interface, displaying this user model in real time. The dashboard can also be used to control the user model and the system's behavior. Finally, we discuss a study in which users conversed with the instrumented system. Our results suggest that users appreciate seeing internal states, which helped them expose biased behavior and increased their sense of control. Participants also made valuable suggestions that point to future directions for both design and machine learning research. The project page and video demo of our TalkTuner system are available at https://bit.ly/talktuner-project-page
△ Less
Submitted 15 June, 2024; v1 submitted 12 June, 2024;
originally announced June 2024.
-
Rainbow ortho-convex 4-sets in k-colored point sets
Authors:
David Flores-Peñaloza,
Mario A. Lopez,
Nestaly Marín,
David Orden
Abstract:
Let $P$ be a $k$-colored set of $n$ points in the plane, $4 \leq k \leq n$. We study the problem of deciding if $P$ contains a subset of four points of different colors such that its Rectilinear Convex Hull has positive area. We provide an $O(n \log n)$-time algorithm for this problem, where the hidden constant does not depend on $k$; then, we prove that this problem has time complexity…
▽ More
Let $P$ be a $k$-colored set of $n$ points in the plane, $4 \leq k \leq n$. We study the problem of deciding if $P$ contains a subset of four points of different colors such that its Rectilinear Convex Hull has positive area. We provide an $O(n \log n)$-time algorithm for this problem, where the hidden constant does not depend on $k$; then, we prove that this problem has time complexity $Ω(n \log n)$ in the algebraic computation tree model. No general position assumptions for $P$ are required.
△ Less
Submitted 9 April, 2024;
originally announced April 2024.
-
Crossing and intersecting families of geometric graphs on point sets
Authors:
José Luis Álvarez-Rebollar,
Jorge Cravioto-Lagos,
Nestaly Marín,
Oriol Solé-Pi,
Jorge Urrutia
Abstract:
Let $S$ be a set of $n$ points in the plane in general position. Two line segments connecting pairs of points of $S$ cross if they have an interior point in common. Two vertex disjoint geometric graphs with vertices in $S$ cross if there are two edges, one from each graph, which cross. A set of vertex disjoint geometric graphs with vertices in $S$ is called mutually crossing if any two of them cro…
▽ More
Let $S$ be a set of $n$ points in the plane in general position. Two line segments connecting pairs of points of $S$ cross if they have an interior point in common. Two vertex disjoint geometric graphs with vertices in $S$ cross if there are two edges, one from each graph, which cross. A set of vertex disjoint geometric graphs with vertices in $S$ is called mutually crossing if any two of them cross.
We show that there exists a constant $c$ such that from any family of $n$ mutually crossing triangles, one can always obtain a family of at least $n^c$ mutually crossing $2$-paths (each of which is the result of deleting an edge from one of the triangles) and then provide an example that implies that $c$ cannot be taken to be larger than $2/3$. For every $n$ we determine the maximum number of crossings that a Hamiltonian cycle on a set of $n$ points might have. Next, we construct a point set whose longest perfect matching contains no crossings. We also consider edges consisting of a horizontal and a vertical line segment joining pairs of points of $S$, which we call elbows, and prove that in any point set $S$ there exists a family of $\lfloor n/4 \rfloor$ vertex disjoint mutually crossing elbows. Additionally, we show a point set that admits no more than $n/3$ mutually crossing elbows.
Finally we study intersecting families of graphs, which are not necessarily vertex disjoint. A set of edge disjoint graphs with vertices in $S$ is called an intersecting family if for any two graphs in the set we can choose an edge in each of them such that they cross. We prove a conjecture by Lara and Rubio-Montiel, namely, that any set $S$ of $n$ points in general position admits a family of intersecting triangles with a quadratic number of elements.
Some other results are obtained throughout this work.
△ Less
Submitted 17 November, 2022;
originally announced November 2022.
-
Optimal Placement of Base Stations in Border Surveillance using Limited Capacity Drones
Authors:
S. Bereg,
J. M. Díaz-Báñez,
M. Haghpanah,
P. Horn,
M. A. Lopez,
N. Marín,
A. Ramírez-Vigueras,
F. Rodríguez,
O. Solé-Pi,
A. Stevens,
J. Urrutia
Abstract:
Imagine an island modeled as a simple polygon $¶$ with $n$ vertices whose coastline we wish to monitor. We consider the problem of building the minimum number of refueling stations along the boundary of $¶$ in such a way that a drone can follow a polygonal route enclosing the island without running out of fuel. A drone can fly a maximum distance $d$ between consecutive stations and is restricted t…
▽ More
Imagine an island modeled as a simple polygon $¶$ with $n$ vertices whose coastline we wish to monitor. We consider the problem of building the minimum number of refueling stations along the boundary of $¶$ in such a way that a drone can follow a polygonal route enclosing the island without running out of fuel. A drone can fly a maximum distance $d$ between consecutive stations and is restricted to move either along the boundary of $¶$ or its exterior (i.e., over water). We present an algorithm that, given $\mathcal P$, finds the locations for a set of refueling stations whose cardinality is at most the optimal plus one. The time complexity of this algorithm is $O(n^2 + \frac{L}{d} n)$, where $L$ is the length of $\mathcal P$. We also present an algorithm that returns an additive $ε$-approximation for the problem of minimizing the fuel capacity required for the drones when we are allowed to place $k$ base stations around the boundary of the island; this algorithm also finds the locations of these refueling stations. Finally, we propose a practical discretization heuristic which, under certain conditions, can be used to certify optimality of the results.
△ Less
Submitted 27 September, 2022;
originally announced September 2022.
-
Ortho-unit polygons can be guarded with at most $\lfloor \frac{n-4}{8} \rfloor$ guards
Authors:
J. M. Díaz-Báñez,
P. Horn,
M. A. Lopez,
N. Marín,
A. Ramírez-Vigueras,
O. Solé-Pi,
A. Stevens,
J. Urrutia
Abstract:
An orthogonal polygon is called an ortho-unit polygon if its vertices have integer coordinates, and all of its edges have length one. In this paper we prove that any ortho-unit polygon with $n \geq 12$ vertices can be guarded with at most $\lfloor \frac{n-4}{8} \rfloor$ guards.
An orthogonal polygon is called an ortho-unit polygon if its vertices have integer coordinates, and all of its edges have length one. In this paper we prove that any ortho-unit polygon with $n \geq 12$ vertices can be guarded with at most $\lfloor \frac{n-4}{8} \rfloor$ guards.
△ Less
Submitted 26 August, 2022;
originally announced August 2022.
-
Tensor-based Collaborative Filtering With Smooth Ratings Scale
Authors:
Nikita Marin,
Elizaveta Makhneva,
Maria Lysyuk,
Vladimir Chernyy,
Ivan Oseledets,
Evgeny Frolov
Abstract:
Conventional collaborative filtering techniques don't take into consideration the effect of discrepancy in users' rating perception. Some users may rarely give 5 stars to items while others almost always assign 5 stars to the chosen item. Even if they had experience with the same items this systematic discrepancy in their evaluation style will lead to the systematic errors in the ability of recomm…
▽ More
Conventional collaborative filtering techniques don't take into consideration the effect of discrepancy in users' rating perception. Some users may rarely give 5 stars to items while others almost always assign 5 stars to the chosen item. Even if they had experience with the same items this systematic discrepancy in their evaluation style will lead to the systematic errors in the ability of recommender system to effectively extract right patterns from data. To mitigate this problem we introduce the ratings' similarity matrix which represents the dependency between different values of ratings on the population level. Hence, if on average the correlations between ratings exist, it is possible to improve the quality of proposed recommendations by off-setting the effect of either shifted down or shifted up users' rates.
△ Less
Submitted 10 May, 2022;
originally announced May 2022.
-
Specificity measures and reference
Authors:
Albert Gatt,
Nicolás Marín,
Gustavo Rivas-Gervilla,
Daniel Sánchez
Abstract:
In this paper we study empirically the validity of measures of referential success for referring expressions involving gradual properties. More specifically, we study the ability of several measures of referential success to predict the success of a user in choosing the right object, given a referring expression. Experimental results indicate that certain fuzzy measures of success are able to pred…
▽ More
In this paper we study empirically the validity of measures of referential success for referring expressions involving gradual properties. More specifically, we study the ability of several measures of referential success to predict the success of a user in choosing the right object, given a referring expression. Experimental results indicate that certain fuzzy measures of success are able to predict human accuracy in reference resolution. Such measures are therefore suitable for the estimation of the success or otherwise of a referring expression produced by a generation algorithm, especially in case the properties in a domain cannot be assumed to have crisp denotations.
△ Less
Submitted 30 September, 2018;
originally announced October 2018.
