-
Can you text what is happening? Integrating pre-trained language encoders into trajectory prediction models for autonomous driving
Authors:
Ali Keysan,
Andreas Look,
Eitan Kosman,
Gonca Gürsun,
Jörg Wagner,
Yu Yao,
Barbara Rakitsch
Abstract:
In autonomous driving tasks, scene understanding is the first step towards predicting the future behavior of the surrounding traffic participants. Yet, how to represent a given scene and extract its features are still open research questions. In this study, we propose a novel text-based representation of traffic scenes and process it with a pre-trained language encoder.
First, we show that text-…
▽ More
In autonomous driving tasks, scene understanding is the first step towards predicting the future behavior of the surrounding traffic participants. Yet, how to represent a given scene and extract its features are still open research questions. In this study, we propose a novel text-based representation of traffic scenes and process it with a pre-trained language encoder.
First, we show that text-based representations, combined with classical rasterized image representations, lead to descriptive scene embeddings. Second, we benchmark our predictions on the nuScenes dataset and show significant improvements compared to baselines. Third, we show in an ablation study that a joint encoder of text and rasterized images outperforms the individual encoders confirming that both representations have their complementary strengths.
△ Less
Submitted 13 September, 2023; v1 submitted 11 September, 2023;
originally announced September 2023.
-
Routing-Aware Partitioning of the Internet Address Space for Server Ranking in CDNs
Authors:
Gonca Gursun
Abstract:
The goal of Content Delivery Networks (CDNs) is to serve content to end-users with high performance. In order to do that, a CDN measures the latency on the paths from its servers to users and then selects a best available server for each user. For large CDNs, monitoring paths from thousands of servers to millions of users is a challenging task due to its size. In this paper, we address this proble…
▽ More
The goal of Content Delivery Networks (CDNs) is to serve content to end-users with high performance. In order to do that, a CDN measures the latency on the paths from its servers to users and then selects a best available server for each user. For large CDNs, monitoring paths from thousands of servers to millions of users is a challenging task due to its size. In this paper, we address this problem and propose a framework to scale the task of path monitoring. Simply stated, the goal of our framework is clustering IP addresses (clients) such that in each cluster the choice of best available server is same (or similar). Then, finding a best available server for one client in a given cluster will be sufficient to assign that server to the rest of the clients in the cluster.
To achieve this goal, first we introduce two distance metrics to compute how similar the server choices of any given two clients. Second, we use a clustering method that is based on interdomain routing information. We evaluate the goodness of our clusters by using the metrics we introduce. We show that there is a strong correlation between the similarity in how two destination clients are routed to in the Internet and the similarity in their server selections. Finally, we show how to choose representative clients from each cluster so that it is sufficient to learn the latencies from the CDN servers to the representative and find a best available server accordingly for the rest of the clients in the same cluster.
△ Less
Submitted 21 October, 2018;
originally announced October 2018.
-
On Spectral Analysis of the Internet Delay Space and Detecting Anomalous Routing Paths
Authors:
Gonca Gürsun
Abstract:
Latency is one of the most critical performance metrics for a wide range of applications. Therefore, it is important to understand the underlying mechanisms that give rise to the observed latency values and diagnose the ones that are unexpectedly high. In this paper, we study the Internet delay space via robust principal component analysis (RPCA). Using RPCA, we show that the delay space, i.e. the…
▽ More
Latency is one of the most critical performance metrics for a wide range of applications. Therefore, it is important to understand the underlying mechanisms that give rise to the observed latency values and diagnose the ones that are unexpectedly high. In this paper, we study the Internet delay space via robust principal component analysis (RPCA). Using RPCA, we show that the delay space, i.e. the matrix of measured round trip times between end hosts, can be decomposed into two components - the expected latency between end hosts with respect to the current state of the Internet and the inflation on the paths between the end hosts. Using this decomposition, first we study the well-known low-dimensionality phenomena of the delay space and ask what properties of the end hosts define the dimensions. Second, using the decomposition, we develop a filtering method to detect the paths which experience unexpected latencies and identify routing anomalies. We show that our filter successfully identifies an anomalous route even when its observed latency is not obviously high in magnitude.
△ Less
Submitted 1 November, 2017;
originally announced November 2017.
