Search

The rapid development of advanced AI agents and the imminent deployment of many instances of these agents will give rise to multi-agent systems of unprecedented complexity. These systems pose novel and under-explored risks. In this report, we provide a structured taxonomy of these risks by identifying three key failure modes (miscoordination, conflict, and collusion) based on agents' incentives, as well as seven key risk factors (information asymmetries, network effects, selection pressures, destabilising dynamics, commitment problems, emergent agency, and multi-agent security) that can underpin them. We highlight several important instances of each risk, as well as promising directions to help mitigate them. By anchoring our analysis in a range of real-world examples and experimental evidence, we illustrate the distinct challenges posed by multi-agent systems and their implications for the safety, governance, and ethics of advanced AI., Comment: Cooperative AI Foundation, Technical Report #1

Non-volatile Memory (NVM) technologies present a promising alternative to traditional volatile memories such as SRAM and DRAM. Due to the limited availability of real NVM devices, simulators play a crucial role in architectural exploration and hardware-software co-design. This tutorial presents a simulation toolchain through four detailed case studies, showcasing its applicability to various domains of system design, including hybrid main-memory and cache, compute-in-memory, and wear-leveling design. These case studies provide the reader with practical insights on customizing the toolchain for their specific research needs. The source code is open-sourced., Comment: DFG Priority Program 2377 - Disruptive Memory Technologies

Machine-learning models are increasingly deployed on resource-constrained embedded systems with strict timing constraints. In such scenarios, the worst-case execution time (WCET) of the models is required to ensure safe operation. Specifically, decision trees are a prominent class of machine-learning models and the main building blocks of tree-based ensemble models (e.g., random forests), which are commonly employed in resource-constrained embedded systems. In this paper, we develop a systematic approach for WCET optimization of decision tree implementations. To this end, we introduce a linear surrogate model that estimates the execution time of individual paths through a decision tree based on the path's length and the number of taken branches. We provide an optimization algorithm that constructively builds a WCET-optimal implementation of a given decision tree with respect to this surrogate model. We experimentally evaluate both the surrogate model and the WCET-optimization algorithm. The evaluation shows that the optimization algorithm improves analytically determined WCET by up to $17\%$ compared to an unoptimized implementation.

Quantum computing could impact various industries, with the automotive industry with many computational challenges, from optimizing supply chains and manufacturing to vehicle engineering, being particularly promising. This chapter investigates state-of-the-art quantum algorithms to enhance efficiency, accuracy, and scalability across the automotive value chain. We explore recent advances in quantum optimization, machine learning, and numerical and chemistry simulations, highlighting their potential and limitations. We identify and discuss key challenges in near-term and fault-tolerant algorithms and their practical use in industrial applications. While quantum algorithms show potential in many application domains, current noisy intermediate-scale quantum hardware limits scale and, thus, business benefits. In the long term, fault-tolerant systems promise theoretical speedups; however, they also require further progress in hardware and software (e.\,g., related to error correction and data loading). We expect that with this progress, significant practical benefits will emerge eventually.

Tensor network algorithms can efficiently simulate complex quantum many-body systems by utilizing knowledge of their structure and entanglement. These methodologies have been adapted recently for solving the Navier-Stokes equations, which describe a spectrum of fluid phenomena, from the aerodynamics of vehicles to weather patterns. Within this quantum-inspired paradigm, velocity is encoded as matrix product states (MPS), effectively harnessing the analogy between interscale correlations of fluid dynamics and entanglement in quantum many-body physics. This particular tensor structure is also called quantics tensor train (QTT). By utilizing NVIDIA's cuQuantum library to perform parallel tensor computations on GPUs, our adaptation speeds up simulations by up to 12.1 times. This allows us to study the algorithm in terms of its applicability, scalability, and performance. By simulating two qualitatively different but commonly encountered 2D flow problems at high Reynolds numbers up to $1\times10^7$ using a fourth-order time stepping scheme, we find that the algorithm has a potential advantage over direct numerical simulations in the turbulent regime as the requirements for grid resolution increase drastically. In addition, we derive the scaling $\chi=\mathcal{O}(\text{poly}(1/\epsilon))$ for the maximum bond dimension $\chi$ of MPS representing turbulent flow fields, with an error $\epsilon$, based on the spectral distribution of turbulent kinetic energy. Our findings motivate further exploration of related quantum algorithms and other tensor network methods.

Steerable needles are minimally invasive devices that enable novel medical procedures by following curved paths to avoid critical anatomical obstacles. Planning algorithms can be used to find a steerable needle motion plan to a target. Deployment typically consists of a physician manually inserting the steerable needle into tissue at the motion plan's start pose and handing off control to a robot, which then autonomously steers it to the target along the plan. The handoff between human and robot is critical for procedure success, as even small deviations from the start pose change the steerable needle's workspace and there is no guarantee that the target will still be reachable. We introduce a metric that evaluates the robustness to such start pose deviations. When measuring this robustness to deviations, we consider the tradeoff between being robust to changes in position versus changes in orientation. We evaluate our metric through simulation in an abstract, a liver, and a lung planning scenario. Our evaluation shows that our metric can be combined with different motion planners and that it efficiently determines large, safe start regions., Comment: 16 pages, 10 figures

Biodegradability is a key feature for the application of polymeric devices in medicine. This study reports an experimental and theoretical study of the degradation of poly(thioether-ester) (PTEe) nanoparticles in aqueous media. The $\alpha$,$\omega$-diene diester derived from vegetable oil, 1,3-propylene diundeca-10-polenoate (Pd10e) was used as monomer in the solvent-free synthesis of Pd10e-based nanoparticles (A-PTEe nanoparticles) via thiol-ene miniemulsion polymerization. The theoretical partition coefficients of A-PTEe and a PTEe based on dianhydro-D-glucityl diundec-10-enoate (DGU) (B-PTEe nanoparticles) were calculated using density functional theory (DFT), in order to compare their degradation behavior. The results showed that A-PTEe is more hydrophilic than B-PTEe, thus indicating the possible faster degradation of the former. The experimental degradation studies showed that, in fact, A-PTEe nanoparticles are faster degraded than B-PTEe, presenting substantial molecular weight decrease, which confirms the theoretical results. The effects of degradation could be observed in the chemical composition and thermal properties of the polymer. Considering its applicability potential as a biomaterial due to its fast degradation behavior, the cytotoxicity of A-PTEe nanoparticles and its degradation products were evaluated. In vitro assays confirmed the biocompatibility of A-PTEe nanoparticles and its degradation products when exposed on fibroblasts and red blood cells. These results suggest A-PTEe nanoparticles can be promising candidates as biobased nanocarriers for biomedical applications., Comment: 27 pages, 7 figures, 1 table

Automated knowledge curation for biomedical ontologies is key to ensure that they remain comprehensive, high-quality and up-to-date. In the era of foundational language models, this study compares and analyzes three NLP paradigms for curation tasks: in-context learning (ICL), fine-tuning (FT), and supervised learning (ML). Using the Chemical Entities of Biological Interest (ChEBI) database as a model ontology, three curation tasks were devised. For ICL, three prompting strategies were employed with GPT-4, GPT-3.5, BioGPT. PubmedBERT was chosen for the FT paradigm. For ML, six embedding models were utilized for training Random Forest and Long-Short Term Memory models. Five setups were designed to assess ML and FT model performance across different data availability scenarios.Datasets for curation tasks included: task 1 (620,386), task 2 (611,430), and task 3 (617,381), maintaining a 50:50 positive versus negative ratio. For ICL models, GPT-4 achieved best accuracy scores of 0.916, 0.766 and 0.874 for tasks 1-3 respectively. In a direct comparison, ML (trained on ~260,000 triples) outperformed ICL in accuracy across all tasks. (accuracy differences: +.11, +.22 and +.17). Fine-tuned PubmedBERT performed similarly to leading ML models in tasks 1 & 2 (F1 differences: -.014 and +.002), but worse in task 3 (-.048). Simulations revealed performance declines in both ML and FT models with smaller and higher imbalanced training data. where ICL (particularly GPT-4) excelled in tasks 1 & 3. GPT-4 excelled in tasks 1 and 3 with less than 6,000 triples, surpassing ML/FT. ICL underperformed ML/FT in task 2.ICL-augmented foundation models can be good assistants for knowledge curation with correct prompting, however, not making ML and FT paradigms obsolete. The latter two require task-specific data to beat ICL. In such cases, ML relies on small pretrained embeddings, minimizing computational demands., Comment: 26 pages, 5 figures, 14 tables

Large language models (LLMs) that do not give consistent answers across contexts are problematic when used for tasks with expectations of consistency, e.g., question-answering, explanations, etc. Our work presents an evaluation benchmark for self-consistency in cases of under-specification where two or more answers can be correct. We conduct a series of behavioral experiments on the OpenAI model suite using an ambiguous integer sequence completion task. We find that average consistency ranges from 67\% to 82\%, far higher than would be predicted if a model's consistency was random, and increases as model capability improves. Furthermore, we show that models tend to maintain self-consistency across a series of robustness checks, including prompting speaker changes and sequence length changes. These results suggest that self-consistency arises as an emergent capability without specifically training for it. Despite this, we find that models are uncalibrated when judging their own consistency, with models displaying both over- and under-confidence. We also propose a nonparametric test for determining from token output distribution whether a model assigns non-trivial probability to alternative answers. Using this test, we find that despite increases in self-consistency, models usually place significant weight on alternative, inconsistent answers. This distribution of probability mass provides evidence that even highly self-consistent models internally compute multiple possible responses., Comment: BlackboxNLP @ EMNLP 2023

Anatomical models of a medical robot's environment can significantly help guide design and development of a new robotic system. These models can be used for benchmarking motion planning algorithms, evaluating controllers, optimizing mechanical design choices, simulating procedures, and even as resources for data generation. Currently, the time-consuming task of generating these environments is repeatedly performed by individual research groups and rarely shared broadly. This not only leads to redundant efforts, but also makes it challenging to compare systems and algorithms accurately. In this work, we present a collection of clinically-relevant anatomical environments for medical robots operating in the lungs. Since anatomical deformation is a fundamental challenge for medical robots operating in the lungs, we describe a way to model respiratory deformation in these environments using patient-derived data. We share the environments and deformation data publicly by adding them to the Medical Robotics Anatomical Dataset (Med-RAD), our public dataset of anatomical environments for medical robots.

Recent model editing techniques promise to mitigate the problem of memorizing false or outdated associations during LLM training. However, we show that these techniques can introduce large unwanted side effects which are not detected by existing specificity benchmarks. We extend the existing CounterFact benchmark to include a dynamic component and dub our benchmark CounterFact+. Additionally, we extend the metrics used for measuring specificity by a principled KL divergence-based metric. We use this improved benchmark to evaluate recent model editing techniques and find that they suffer from low specificity. Our findings highlight the need for improved specificity benchmarks that identify and prevent unwanted side effects., Comment: To be published in ACL Findings 2023; for code see https://github.com/apartresearch/specificityplus; for a homepage see https://specificityplus.apartresearch.com/; updated Figures to uniform style

The simulation of many industrially relevant physical processes can be executed up to exponentially faster using quantum algorithms. However, this speedup can only be leveraged if the data input and output of the simulation can be implemented efficiently. While we show that recent advancements for optimal state preparation can effectively solve the problem of data input at a moderate cost of ancillary qubits in many cases, the output problem can provably not be solved efficiently in general. By acknowledging that many simulation problems arise only as a subproblem of a larger optimization problem in many practical applications however, we identify and define a class of practically relevant problems that does not suffer from the output problem: Quantum Simulation-based Optimization (QuSO). QuSO represents optimization problems whose objective function and/or constraints depend on summary statistic information on the result of a simulation, i.e., information that can be efficiently extracted from a quantum state vector. In this article, we focus on the LinQuSO subclass of QuSO, which is characterized by the linearity of the simulation problem, i.e., the simulation problem can be formulated as a system of linear equations. By cleverly combining the quantum singular value transformation (QSVT) with the quantum approximate optimization algorithm (QAOA), we prove that a large subgroup of LinQuSO problems can be solved with up to exponential quantum speedups with regards to their simulation component. Finally, we present two practically relevant use cases that fall within this subgroup of QuSO problems., Comment: 24 pages, 12 figure, completely refactored and formalized version with key new isights

Generative adversarial networks (GANs) have shown remarkable success in generating realistic images and are increasingly used in medical imaging for image-to-image translation tasks. However, GANs tend to suffer from a frequency bias towards low frequencies, which can lead to the removal of important structures in the generated images. To address this issue, we propose a novel frequency-aware image-to-image translation framework based on the supervised RegGAN approach, which we call fRegGAN. The framework employs a K-space loss to regularize the frequency content of the generated images and incorporates well-known properties of MRI K-space geometry to guide the network training process. By combine our method with the RegGAN approach, we can mitigate the effect of training with misaligned data and frequency bias at the same time. We evaluate our method on the public BraTS dataset and outperform the baseline methods in terms of both quantitative and qualitative metrics when synthesizing T2-weighted from T1-weighted MR images. Detailed ablation studies are provided to understand the effect of each modification on the final performance. The proposed method is a step towards improving the performance of image-to-image translation and synthesis in the medical domain and shows promise for other applications in the field of image processing and generation.

The use of needles to access sites within organs is fundamental to many interventional medical procedures both for diagnosis and treatment. Safe and accurate navigation of a needle through living tissue to an intra-tissue target is currently often challenging or infeasible due to the presence of anatomical obstacles in the tissue, high levels of uncertainty, and natural tissue motion (e.g., due to breathing). Medical robots capable of automating needle-based procedures in vivo have the potential to overcome these challenges and enable an enhanced level of patient care and safety. In this paper, we show the first medical robot that autonomously navigates a needle inside living tissue around anatomical obstacles to an intra-tissue target. Our system leverages an aiming device and a laser-patterned highly flexible steerable needle, a type of needle capable of maneuvering along curvilinear trajectories to avoid obstacles. The autonomous robot accounts for anatomical obstacles and uncertainty in living tissue/needle interaction with replanning and control and accounts for respiratory motion by defining safe insertion time windows during the breathing cycle. We apply the system to lung biopsy, which is critical in the diagnosis of lung cancer, the leading cause of cancer-related death in the United States. We demonstrate successful performance of our system in multiple in vivo porcine studies and also demonstrate that our approach leveraging autonomous needle steering outperforms a standard manual clinical technique for lung nodule access., Comment: 22 pages, 6 figures

The most interesting words in scientific texts will often be novel or rare. This presents a challenge for scientific word embedding models to determine quality embedding vectors for useful terms that are infrequent or newly emerging. We demonstrate how \gls{lsi} can address this problem by imputing embeddings for domain-specific words from up-to-date knowledge graphs while otherwise preserving the original word embedding model. We use the MeSH knowledge graph to impute embedding vectors for biomedical terminology without retraining and evaluate the resulting embedding model on a domain-specific word-pair similarity task. We show that LSI can produce reliable embedding vectors for rare and OOV terms in the biomedical domain., Comment: Accepted for the Workshop on Information Extraction from Scientific Publications at AACL-IJCNLP 2022