arXiv subjects

Johannes Fischer

Publications and source records attributed to Johannes Fischer.

At least 19 recordsLinked to original sources

arXiv2026

Average-Case Optimal Encodings and Efficient Worst-Case Indices for Element Distinctness Queries

We study the data structure version of the \emph{element distinctness problem}: preprocess an array of $n$ elements from an alphabet of size $σ$ to answer \textsc{All-Distinct} queries, asking whether a given range contains only distinct elements. We first focus on \emph{uniformly random arrays}: in the encoding model, where access to the input at query time is not allowed, we prove a lower bound on the expected space; for instance, the lower bound is $n$, $1.3627n$, $1.5153n$, $1.5824n$ bits for $σ= 2,3,4,5$, and approximately $n\sqrt{π/(2σ)}\,\logσ$ bits for $σ=ω(1)$. We complement this by designing different average-case optimal encodings, supporting \textsc{All-Distinct} queries in worst-case time $O(1)$, $o(\log^{2}{\log{n}})$, or $O(\log\log{n})$ depending on $σ$, and $O(1)$ expected time for any $σ= ω(1)$. We then switch to worst-case (non-random) arrays: in the indexing model, where access to the input is allowed, we prove a cell-probe space-time tradeoff lower bound showing that any index using $n/b$ bits must have $Ω(b/\log{b})$ query time. We conclude by presenting a simple index almost matching this lower bound.

cs.DS↗

arXiv2025

Safety Reinforced Model Predictive Control (SRMPC): Improving MPC with Reinforcement Learning for Motion Planning in Autonomous Driving

Model predictive control (MPC) is widely used for motion planning, particularly in autonomous driving. Real-time capability of the planner requires utilizing convex approximation of optimal control problems (OCPs) for the planner. However, such approximations confine the solution to a subspace, which might not contain the global optimum. To address this, we propose using safe reinforcement learning (SRL) to obtain a new and safe reference trajectory within MPC. By employing a learning-based approach, the MPC can explore solutions beyond the close neighborhood of the previous one, potentially finding global optima. We incorporate constrained reinforcement learning (CRL) to ensure safety in automated driving, using a handcrafted energy function-based safety index as the constraint objective to model safe and unsafe regions. Our approach utilizes a state-dependent Lagrangian multiplier, learned concurrently with the safe policy, to solve the CRL problem. Through experimentation in a highway scenario, we demonstrate the superiority of our approach over both MPC and SRL in terms of safety and performance measures.

cs.RO↗

arXiv2025

RLZ-r and LZ-End-r: Enhancing Move-r

In pattern matching on strings, a locate query asks for an enumeration of all the occurrences of a given pattern in a given text. The r-index [Gagie et al., 2018] is a recently presented compressed self index that stores the text and auxiliary information in compressed space. With some modifications, locate queries can be answered in optimal time [Nishimoto & Tabei, 2021], which has recently been proven relevant in practice in the form of Move-r [Bertram et al., 2024]. However, there remains the practical bottleneck of evaluating function $Φ$ for every occurrence to report. This motivates enhancing the index by a compressed representation of the suffix array featuring efficient random access, trading off space for faster answering of locate queries [Puglisi & Zhukova, 2021]. In this work, we build upon this idea considering two suitable compression schemes: Relative Lempel-Ziv [Kuruppu et al., 2010], improving the work by Puglisi and Zhukova, and LZ-End [Kreft & Navarro, 2010], introducing a different trade-off where compression is better than for Relative Lempel-Ziv at the cost of slower access times. We enhance both the r-index and Move-r by the compressed suffix arrays and evaluate locate query performance in an experiment. We show that locate queries can be sped up considerably in both the r-index and Move-r, especially if the queried pattern has many occurrences. The choice between two different compression schemes offers new trade-offs regarding index size versus query performance.

cs.DS↗

arXiv2025

Automated Detection of Salvin's Albatrosses: Improving Deep Learning Tools for Aerial Wildlife Surveys

Recent advancements in deep learning and aerial imaging have transformed wildlife monitoring, enabling researchers to survey wildlife populations at unprecedented scales. Unmanned Aerial Vehicles (UAVs) provide a cost-effective means of capturing high-resolution imagery, particularly for monitoring densely populated seabird colonies. In this study, we assess the performance of a general-purpose avian detection model, BirdDetector, in estimating the breeding population of Salvin's albatross (Thalassarche salvini) on the Bounty Islands, New Zealand. Using drone-derived imagery, we evaluate the model's effectiveness in both zero-shot and fine-tuned settings, incorporating enhanced inference techniques and stronger augmentation methods. Our findings indicate that while applying the model in a zero-shot setting offers a strong baseline, fine-tuning with annotations from the target domain and stronger image augmentation leads to marked improvements in detection accuracy. These results highlight the potential of leveraging pre-trained deep-learning models for species-specific monitoring in remote and challenging environments.

cs.CV↗

arXiv2024

ConstrainedZero: Chance-Constrained POMDP Planning using Learned Probabilistic Failure Surrogates and Adaptive Safety Constraints

To plan safely in uncertain environments, agents must balance utility with safety constraints. Safe planning problems can be modeled as a chance-constrained partially observable Markov decision process (CC-POMDP) and solutions often use expensive rollouts or heuristics to estimate the optimal value and action-selection policy. This work introduces the ConstrainedZero policy iteration algorithm that solves CC-POMDPs in belief space by learning neural network approximations of the optimal value and policy with an additional network head that estimates the failure probability given a belief. This failure probability guides safe action selection during online Monte Carlo tree search (MCTS). To avoid overemphasizing search based on the failure estimates, we introduce $Δ$-MCTS, which uses adaptive conformal inference to update the failure threshold during planning. The approach is tested on a safety-critical POMDP benchmark, an aircraft collision avoidance system, and the sustainability problem of safe CO$_2$ storage. Results show that by separating safety constraints from the objective we can achieve a target level of safety without optimizing the balance between rewards and costs.

cs.AI↗

arXiv2024

PITA: Physics-Informed Trajectory Autoencoder

Validating robotic systems in safety-critical appli-cations requires testing in many scenarios including rare edgecases that are unlikely to occur, requiring to complement real-world testing with testing in simulation. Generative models canbe used to augment real-world datasets with generated data toproduce edge case scenarios by sampling in a learned latentspace. Autoencoders can learn said latent representation for aspecific domain by learning to reconstruct the input data froma lower-dimensional intermediate representation. However, theresulting trajectories are not necessarily physically plausible, butinstead typically contain noise that is not present in the inputtrajectory. To resolve this issue, we propose the novel Physics-Informed Trajectory Autoencoder (PITA) architecture, whichincorporates a physical dynamics model into the loss functionof the autoencoder. This results in smooth trajectories that notonly reconstruct the input trajectory but also adhere to thephysical model. We evaluate PITA on a real-world dataset ofvehicle trajectories and compare its performance to a normalautoencoder and a state-of-the-art action-space autoencoder.

cs.LG↗

arXiv2023

Launch Power Optimization for Dynamic Elastic Optical Networks over C+L Bands

We propose an algorithm for calculating the optimum launch power over the entire C+L bands by maximizing the cumulative link GSNR of a channel plan built upon multiple modulation formats, with application to dynamic EONs. Exact last-fit spectrum assignment proves to outperform exact first-fit in terms of average GSNR at arrival time.

eess.SP↗

arXiv2023

SHAIL: Safety-Aware Hierarchical Adversarial Imitation Learning for Autonomous Driving in Urban Environments

Designing a safe and human-like decision-making system for an autonomous vehicle is a challenging task. Generative imitation learning is one possible approach for automating policy-building by leveraging both real-world and simulated decisions. Previous work that applies generative imitation learning to autonomous driving policies focuses on learning a low-level controller for simple settings. However, to scale to complex settings, many autonomous driving systems combine fixed, safe, optimization-based low-level controllers with high-level decision-making logic that selects the appropriate task and associated controller. In this paper, we attempt to bridge this gap in complexity by employing Safety-Aware Hierarchical Adversarial Imitation Learning (SHAIL), a method for learning a high-level policy that selects from a set of low-level controller instances in a way that imitates low-level driving data on-policy. We introduce an urban roundabout simulator that controls non-ego vehicles using real data from the Interaction dataset. We then demonstrate empirically that even with simple controller options, our approach can produce better behavior than previous approaches in driver imitation that have difficulty scaling to complex environments. Our implementation is available at https://github.com/sisl/InteractionImitation.

cs.RO↗

arXiv2023

Pinching Azumaya algebras

We show, that for a morphism of schemes from X to Y, that is a finite modification in finitely many closed points, a cohomological Brauer class on Y is represented by an Azumaya algebra if its pullback to X is represented by an Azumaya algebra. Part of the proof uses an extension of a result by Ferrand, on pinching of finite locally free sheaves, to Azumaya algebras.

math.AG↗

arXiv2023

Sliding Window String Indexing in Streams

Given a string $S$ over an alphabet $Σ$, the 'string indexing problem' is to preprocess $S$ to subsequently support efficient pattern matching queries, i.e., given a pattern string $P$ report all the occurrences of $P$ in $S$. In this paper we study the 'streaming sliding window string indexing problem'. Here the string $S$ arrives as a stream, one character at a time, and the goal is to maintain an index of the last $w$ characters, called the 'window', for a specified parameter $w$. At any point in time a pattern matching query for a pattern $P$ may arrive, also streamed one character at a time, and all occurrences of $P$ within the current window must be returned. The streaming sliding window string indexing problem naturally captures scenarios where we want to index the most recent data (i.e. the window) of a stream while supporting efficient pattern matching. Our main result is a simple $O(w)$ space data structure that uses $O(\log w)$ time with high probability to process each character from both the input string $S$ and the pattern string $P$. Reporting each occurrence from $P$ uses additional constant time per reported occurrence. Compared to previous work in similar scenarios this result is the first to achieve an efficient worst-case time per character from the input stream. We also consider a delayed variant of the problem, where a query may be answered at any point within the next $δ$ characters that arrive from either stream. We present an $O(w + δ)$ space data structure for this problem that improves the above time bounds to $O(\log(w/δ))$. In particular, for a delay of $δ= εw$ we obtain an $O(w)$ space data structure with constant time processing per character. The key idea to achieve our result is a novel and simple hierarchical structure of suffix trees of independent interest, inspired by the classic log-structured merge trees.

cs.DS↗

arXiv2021

Minimizing Safety Interference for Safe and Comfortable Automated Driving with Distributional Reinforcement Learning

Despite recent advances in reinforcement learning (RL), its application in safety critical domains like autonomous vehicles is still challenging. Although punishing RL agents for risky situations can help to learn safe policies, it may also lead to highly conservative behavior. In this paper, we propose a distributional RL framework in order to learn adaptive policies that can tune their level of conservativity at run-time based on the desired comfort and utility. Using a proactive safety verification approach, the proposed framework can guarantee that actions generated from RL are fail-safe according to the worst-case assumptions. Concurrently, the policy is encouraged to minimize safety interference and generate more comfortable behavior. We trained and evaluated the proposed approach and baseline policies using a high level simulator with a variety of randomized scenarios including several corner cases which rarely happen in reality but are very crucial. In light of our experiments, the behavior of policies learned using distributional RL can be adaptive at run-time and robust to the environment uncertainty. Quantitatively, the learned distributional RL agent drives in average 8 seconds faster than the normal DQN policy and requires 83\% less safety interference compared to the rule-based policy with slightly increasing the average crossing time. We also study sensitivity of the learned policy in environments with higher perception noise and show that our algorithm learns policies that can still drive reliable when the perception noise is two times higher than the training configuration for automated merging and crossing at occluded intersections.

cs.RO↗

arXiv2021

Compression with the tudocomp Framework

We present a framework facilitating the implementation and comparison of text compression algorithms. We evaluate its features by a case study on two novel compression algorithms based on the Lempel-Ziv compression schemes that perform well on highly repetitive texts.

cs.DS↗

arXiv2021

Engineering Predecessor Data Structures for Dynamic Integer Sets

We present highly optimized data structures for the dynamic predecessor problem, where the task is to maintain a set $S$ of $w$-bit numbers under insertions, deletions, and predecessor queries (return the largest element in $S$ no larger than a given key). The problem of finding predecessors can be viewed as a generalized form of the membership problem, or as a simple version of the nearest neighbour problem. It lies at the core of various real-world problems such as internet routing. In this work, we engineer (1) a simple implementation of the idea of universe reduction, similar to van-Emde-Boas trees (2) variants of y-fast tries [Willard, IPL'83], and (3) B-trees with different strategies for organizing the keys contained in the nodes, including an implementation of dynamic fusion nodes [Pǎtraşcu and Thorup, FOCS'14]. We implement our data structures for $w=32,40,64$, which covers most typical scenarios. Our data structures finish workloads faster than previous approaches while being significantly more space-efficient, e.g., they clearly outperform standard implementations of the STL by finishing up to four times as fast using less than a third of the memory. Our tests also provide more general insights on data structure design, such as how small sets should be stored and handled and if and when new CPU instructions such as advanced vector extensions pay off.

cs.DS↗

arXiv2021

Linear Time Runs over General Ordered Alphabets

A run in a string is a maximal periodic substring. For example, the string $\texttt{bananatree}$ contains the runs $\texttt{anana} = (\texttt{an})^{3/2}$ and $\texttt{ee} = \texttt{e}^2$. There are less than $n$ runs in any length-$n$ string, and computing all runs for a string over a linearly-sortable alphabet takes $\mathcal{O}(n)$ time (Bannai et al., SODA 2015). Kosolobov conjectured that there also exists a linear time runs algorithm for general ordered alphabets (Inf. Process. Lett. 2016). The conjecture was almost proven by Crochemore et al., who presented an $\mathcal{O}(nα(n))$ time algorithm (where $α(n)$ is the extremely slowly growing inverse Ackermann function). We show how to achieve $\mathcal{O}(n)$ time by exploiting combinatorial properties of the Lyndon array, thus proving Kosolobov's conjecture.

cs.DS↗

arXiv2020

LCP-Aware Parallel String Sorting

When lexicographically sorting strings, it is not always necessary to inspect all symbols. For example, the lexicographical rank of "europar" amongst the strings "eureka", "eurasia", and "excells" only depends on its so called relevant prefix "euro". The distinguishing prefix size $D$ of a set of strings is the number of symbols that actually need to be inspected to establish the lexicographical ordering of all strings. Efficient string sorters should be $D$-aware, i.e. their complexity should depend on $D$ rather than on the total number $N$ of all symbols in all strings. While there are many $D$-aware sorters in the sequential setting, there appear to be no such results in the PRAM model. We propose a framework yielding a $D$-aware modification of any existing PRAM string sorter. The derived algorithms are work-optimal with respect to their original counterpart: If the original algorithm requires $O(w(N))$ work, the derived one requires $O(w(D))$ work. The execution time increases only by a small factor that is logarithmic in the length of the longest relevant prefix. Our framework universally works for deterministic and randomized algorithms in all variations of the PRAM model, such that future improvements in ($D$-unaware) parallel string sorting will directly result in improvements in $D$-aware parallel string sorting.

cs.DS↗

arXiv2020

Towards an Interoperable Ecosystem of AI and LT Platforms: A Roadmap for the Implementation of Different Levels of Interoperability

With regard to the wider area of AI/LT platform interoperability, we concentrate on two core aspects: (1) cross-platform search and discovery of resources and services; (2) composition of cross-platform service workflows. We devise five different levels (of increasing complexity) of platform interoperability that we suggest to implement in a wider federation of AI/LT platforms. We illustrate the approach using the five emerging AI/LT platforms AI4EU, ELG, Lynx, QURATOR and SPEAKER.

cs.CL↗

arXiv2019

Space Efficient Construction of Lyndon Arrays in Linear Time

We present the first linear time algorithm to construct the $2n$-bit version of the Lyndon array for a string of length $n$ using only $o(n)$ bits of working space. A simpler variant of this algorithm computes the plain ($n\lg n$-bit) version of the Lyndon array using only $\mathcal{O}(1)$ words of additional working space. All previous algorithms are either not linear, or use at least $n\lg n$ bits of additional working space. Also in practice, our new algorithms outperform the previous best ones by an order of magnitude, both in terms of time and space.

cs.DS↗

arXiv2019

Bidirectional Text Compression in External Memory

Bidirectional compression algorithms work by substituting repeated substrings by references that, unlike in the famous LZ77-scheme, can point to either direction. We present such an algorithm that is particularly suited for an external memory implementation. We evaluate it experimentally on large data sets of size up to 128 GiB (using only 16 GiB of RAM) and show that it is significantly faster than all known LZ77 compressors, while producing a roughly similar number of factors. We also introduce an external memory decompressor for texts compressed with any uni- or bidirectional compression scheme.

cs.DS↗