26838 research outputs found
Sort by
Diagnosing Bias in Recommender Systems - Elliot
No full textIn this project, we experiment with detecting bias in recommender systems, using the framework Elliot, and testing the DMF algorithm
The Queue Automaton Revisited
Get PDFWe consider the computational model of the Queue Automaton. An old result is that the deterministic queue automaton is equally expressive as the Turing machine. We introduced the Reactive Turing Machine, enhancing the Turing machine with a notion of interaction. The Reactive Turing Machine defines all executable processes. In this paper, we prove that the non-deterministic queue automaton is equally expressive as the Reactive Turing Machine. Together with finite automata, pushdown automata and parallel pushdown automata, queue automata form a nice hierarchy of executable processes, with stacks, bags and queues as central elements
Review argumentation at scale
Get PDFProduct reviews represent a valuable source of information for both (potential) customers and sellers. Usually, reviews come in pairs (score, motivation), where the motivation is a piece of unstructured text explaining the score given to a product. For reviews, this setting is ideal to combine a quantitative assessment of a product with a qualitative explanation. Aggregating the numerical scores might be uninformative while parsing large quantities of text might be challenging.
Automated argument analysis can help in this process, and we previously developed an argument-based quality analysis pipeline that helps identify the most significant items from a corpus of reviews. Given that the pipeline is effective but time-consuming, this work sets out to improve its computational efficiency. Next to optimisation by conventional methods, we investigate the effect of reducing the number of text chunks that are used to build the argumentation graph.
We find that conventional methods significantly improve the computation time, which allows us to analyse much larger datasets of real-world reviews. When the number of tokens is scaled down, accuracy remains similar compared to the original version of the pipeline. However, we find that this does not necessarily result in a computation time reduction
Real-time assistance in suicide prevention helplines using a deep learning-based recommender system: A randomized controlled trial
Get PDFObjective: To evaluate the effectiveness and usability of an AI-assisted tool in providing real-time assistance to
counselors during suicide prevention helpline conversations.
Methods: In this RCT, the intervention group used an AI-assisted tool, which generated suggestions based on
sentence embeddings (i.e. BERT) from previous successful counseling sessions. Cosine similarity was used to
present the top 5 chat situation to the counsellors. The control group did not have access to the tool (care as
usual). Both groups completed a questionnaire assessing their self-efficacy at the end of each shift. Counselors’
usage of the tool was evaluated by measuring frequency, duration and content of interactions.
Results: In total, 48 counselors participated in the experiment: 27 counselors in the experimental condition and 21
counselors in the control condition. Together they rated 188 shifts. No significant difference in self-efficacy was
observed between the two groups (p=0.36). However, counselors that used the AI-assisted tool had marginally
lower response time and used the tool more often during conversations that had a longer duration. A deeper
analysis of usage showed that the tool was frequently used in inappropriate situations, e.g. after the counselor
had already provided a response to the help-seeker, defeating the purpose of the information. When the tool was
employed appropriately (64 conversations), it provided usable information in 53 conversations (83%). However,
counselors used the tool less frequently at optimal moments, indicating their potential lack of prficiency with
using AI-assisted tools during helpline conversations or initial trust issues with the system.
Conclusion: The study demonstrates benfits and pitfalls of integrating AI-assisted tools in suicide prevention for
improving counselor support. Despite the lack of significant impact on self-efficacy, the support tool provided
usable suggestions and the frequent use during long conversations suggests counsellors may wish to use the tool
in complex or challenging interactions
Towards didactics-driven development of educational games
Get PDFEducational games offer an effective means to enhance interactive
learning experiences. However, developing high quality educational
games is difficult. In particular, integrating didactic goals into a
game’s design, and verifying the learning outcomes is a complex
iterative process. Due to a lack of control over these concerns, fully
developed games may lack the intended educational value.
We aim to improve educational game development by structuring
and partly automating this process. We propose Didactics-Driven
Development, a novel framework for keeping didactic goals and
concerns in focus throughout the development process. By making
didactic concerns explicit, it enables testing interaction patterns
against opportunities for learning. We discuss how the approach
has been applied to three case studies and how this has informed
ongoing development of the framework
Deep learning for landmark detection, segmentation, and multi-objective deformable registration in medical imaging
No full textCervical cancer affects about half a million women globally every year. The treatment of cervical cancer with the aim of healing mainly consists of surgery, radiation treatment, or a combination of radiation treatment with chemotherapy or hyperthermia. Radiation treatment is a type of treatment wherein a high dose of ionizing radiation is used to kill the tumor cells. The radiation dose is usually delivered in the form of External Beam Radiation Treatment (EBRT) with a linear accelerator followed by internal radiation treatment (brachytherapy) during which a small radioactive source is passed through an applicator and needles that are placed temporarily nearby the cervix. EBRT typically spans several weeks with daily sessions (often referred to as fractions), whereas brachytherapy typically consists of three or four fractions based on one to three implantations. The aim of the radiation treatment is to provide effective radiation to kill the tumor cells while sparing the nearby healthy tissue or Organs At Risk (OARs) as much as possible. This is achieved by treatment planning following the contouring of target volumes and OARs, on medical imaging scans, which typically are Computed Tomography (CT) and/orMagnetic Resonance Imaging (MRI)
VCrypt: Leveraging Vectorized and Compressed Execution for Client-side Encryption
Get PDFVCrypt is a novel extension on DuckDB that enables fine-grained client-side [en/de]cryption in a performance- and storage-efficient manner, by exploiting columnar compression as well as vectorized and compressed execution. We designed VCrypt such that in analytical queries, typically (i) data can be encrypted and decrypted batch-at-a-time instead of value-at-a-time, and (ii) the extra storage for cryptographic nonces gets compressed away. We also demonstrate the use of VCrypt inside MotherDuck, leveraging its hybrid processing model that evaluates SQL queries partly on a client DuckDB and partly on a cloud DuckDB, to achieve secure hybrid execution. This provides security even if the cloud server is untrusted, by forcing the [en/de]cryption of sensitive data to happen only client-side, while still allowing useful cloud-side work like filters and joins
Towards a modern LLL implementation
Get PDFWe propose BLASter, a proof of concept LLL implementation
that demonstrates the practicality of multiple theoretical improvements.
The implementation uses the segmentation strategy from Neumaier–
Stehlé (ISSAC 2016), parallelism and Seysen’s reduction that was pro-
posed by Kirchner–Espitau–Fouque (CRYPTO 2021) and implemented
in OptLLL, and the BLAS library for linear algebra operations. It con-
sists of only 1000 significant lines of C++ and Python code, and is made
publicly available.
For q-ary lattices that fplll can handle without multiprecision (dimen-
sion < 180), BLASter is considerably faster than fplll, OptLLL and Ryan–
Heninger’s flatter (CRYPTO 2023), without degrading output reduction
quality. Thanks to Seysen’s reduction it can further handle larger dimen-
sion without resorting to multiprecision, making it more than 10x faster
than flatter and OptLLL, and 100x faster than fplll in dimensions 256 to
1024.
It further includes segmented BKZ and segmented deep-LLL variants.
The latter provides bases as good as BKZ-15 and has a runtime that is
only a couple of times more than our LLL baseline.
This remains a proof of concept: the effective use of higher precision —
which is needed to handle all lattices — has further obstacles and is left
for future work. Still, this work contains many lessons learned, and is
meant to motivate and guide the development of a robust and modern
lattice reduction library, which shall be much faster than fplll
Studying exploration in RL: An optimal transport analysis of occupancy measure trajectories
No full textThe rising successes of RL are propelled by combining smart algorithmic strategies and deep architectures to optimize the distribution of returns and visitations over the state-action space. A quantitative framework to compare the learning processes of these eclectic RL algorithms is currently absent but desired in practice. We address this gap by representing the learning process of an RL algorithm as a sequence of policies generated during training, and then studying the policy trajectory induced in the manifold of state-action occupancy measures. Using an optimal transport-based metric, we measure the length of the paths induced by the policy sequence yielded by an RL algorithm between an initial policy and a final optimal policy. Hence, we first define the Effort of Sequential Learning (ESL). ESL quantifies the relative distance that an RL algorithm travels compared to the shortest path from the initial to the optimal policy. Furthermore, we connect the dynamics of policies in the occupancy measure space and regret (another metric to understand the suboptimality of an RL algorithm), by defining the Optimal Movement Ratio (OMR). OMR assesses the fraction of movements in the occupancy measure space that effectively reduce an analogue of regret. Finally, we derive approximation guarantees to estimate ESL and OMR with a finite number of samples and without access to an optimal policy. Through empirical analyses across various environments and algorithms, we demonstrate that ESL and OMR provide insights into the exploration processes of RL algorithms and the hardness of different tasks in discrete and continuous MDPs
Distinguishing graph states by the properties of their marginals
Get PDFGraph states are a class of multipartite entangled quantum states that are ubiquitous in quantum information. We study equivalence relations between graph states under local unitaries (LU) to obtain distinguishing methods both in local and in networked settings. Based on the marginal structure of graph states, we introduce a family of easy-to-compute LU invariants. We show that these invariants uniquely identify the entanglement classes of every graph state up to eight qubits and discuss their reliability for larger numbers of qubits. To handle larger graphs, we generalize tools to test for local Clifford (LC) equivalence of graph states that work by condensing large graphs into smaller graphs. In turn, we show that statements on the equivalence of these smaller graphs (which are easier to compute) can be used to infer statements on the equivalence of the original, larger graphs. We analyze LU equivalence in two key settings, with and without allowing for the permutation of qubits. We identify entanglement classes, whose marginal structure does not allow us to distinguish them. As a result, we increase the bound on the number of qubits where the LU-LC conjecture holds from 8 to 10 qubits in the setting where qubit permutations are allowed