Search

Causal learning is the cognitive process of developing the capability of making causal inferences based on available information, often guided by normative principles. This process is prone to errors and biases, such as the illusion of causality, in which people perceive a causal relationship between two variables despite lacking supporting evidence. This cognitive bias has been proposed to underlie many societal problems, including social prejudice, stereotype formation, misinformation, and superstitious thinking. In this research, we investigate whether large language models (LLMs) develop causal illusions, both in real-world and controlled laboratory contexts of causal learning and inference. To this end, we built a dataset of over 2K samples including purely correlational cases, situations with null contingency, and cases where temporal information excludes the possibility of causality by placing the potential effect before the cause. We then prompted the models to make statements or answer causal questions to evaluate their tendencies to infer causation erroneously in these structured settings. Our findings show a strong presence of causal illusion bias in LLMs. Specifically, in open-ended generation tasks involving spurious correlations, the models displayed bias at levels comparable to, or even lower than, those observed in similar studies on human subjects. However, when faced with null-contingency scenarios or temporal cues that negate causal relationships, where it was required to respond on a 0-100 scale, the models exhibited significantly higher bias. These findings suggest that the models have not uniformly, consistently, or reliably internalized the normative principles essential for accurate causal learning., Comment: 15 pages, 6 figures

Previous work has shown that training "helpful-only" LLMs with reinforcement learning on a curriculum of gameable environments can lead models to generalize to egregious specification gaming, such as editing their own reward function or modifying task checklists to appear more successful. We show that gpt-4o, gpt-4o-mini, o1-preview, and o1-mini - frontier models trained to be helpful, harmless, and honest - can engage in specification gaming without training on a curriculum of tasks, purely from in-context iterative reflection (which we call in-context reinforcement learning, "ICRL"). We also show that using ICRL to generate highly-rewarded outputs for expert iteration (compared to the standard expert iteration reinforcement learning algorithm) may increase gpt-4o-mini's propensity to learn specification-gaming policies, generalizing (in very rare cases) to the most egregious strategy where gpt-4o-mini edits its own reward function. Our results point toward the strong ability of in-context reflection to discover rare specification-gaming strategies that models might not exhibit zero-shot or with normal training, highlighting the need for caution when relying on alignment of LLMs in zero-shot settings., Comment: 20 pages, 9 figures

As AI models become ever more complex and intertwined in humans' daily lives, greater levels of interactivity of explainable AI (XAI) methods are needed. In this paper, we propose the use of belief change theory as a formal foundation for operators that model the incorporation of new information, i.e. user feedback in interactive XAI, to logical representations of data-driven classifiers. We argue that this type of formalisation provides a framework and a methodology to develop interactive explanations in a principled manner, providing warranted behaviour and favouring transparency and accountability of such interactions. Concretely, we first define a novel, logic-based formalism to represent explanatory information shared between humans and machines. We then consider real world scenarios for interactive XAI, with different prioritisations of new and existing knowledge, where our formalism may be instantiated. Finally, we analyse a core set of belief change postulates, discussing their suitability for our real world settings and pointing to particular challenges that may require the relaxation or reinterpretation of some of the theoretical assumptions underlying existing operators., Comment: 9 pages. To be published at KR 2024

Effective adaptation to distribution shifts in training data is pivotal for sustaining robustness in neural networks, especially when removing specific biases or outdated information, a process known as machine unlearning. Traditional approaches typically assume that data variations are random, which makes it difficult to adjust the model parameters accurately to remove patterns and characteristics from unlearned data. In this work, we present Unlearning Information Bottleneck (UIB), a novel information-theoretic framework designed to enhance the process of machine unlearning that effectively leverages the influence of systematic patterns and biases for parameter adjustment. By proposing a variational upper bound, we recalibrate the model parameters through a dynamic prior that integrates changes in data distribution with an affordable computational cost, allowing efficient and accurate removal of outdated or unwanted data patterns and biases. Our experiments across various datasets, models, and unlearning methods demonstrate that our approach effectively removes systematic patterns and biases while maintaining the performance of models post-unlearning.

Weakly interacting massive particles (WIMPs) are well-motivated candidates for dark matter. One signature of galactic WIMPs is the annual modulation expected in a detector's interaction rate, which arises from Earth's revolution around the Sun. Over two decades, the DAMA/LIBRA experiment has observed such modulation with 250 kg of NaI(Tl) scintillators, in accordance with WIMP expectations but inconsistent with the negative results of other experiments. The signal depends on the target material, so to validate or refute the DAMA result, the experiment must be replicated using the same material. This is the goal of the ANAIS-112 experiment, currently underway since August 2017 with 112.5 kg of NaI(Tl). In this work, we present a reanalysis of three years of data employing an improved analysis chain to enhance the experimental sensitivity. The results presented here are consistent with the absence of modulation and inconsistent with DAMA's observation at nearly 3$\sigma$ confidence level, with the potential to reach a 5$\sigma$ level within 8 years from the beginning of the data collection. Additionally, we explore the impact of different scintillation quenching factors in the comparison between ANAIS-112 and DAMA/LIBRA., Comment: 16 pages, 10 figures, 1 table

Preferences are a pivotal component in practical reasoning, especially in tasks that involve decision-making over different options or courses of action that could be pursued. In this work, we focus on repairing and querying inconsistent knowledge bases in the form of graph databases, which involves finding a way to solve conflicts in the knowledge base and considering answers that are entailed from every possible repair, respectively. Without a priori domain knowledge, all possible repairs are equally preferred. Though that may be adequate for some settings, it seems reasonable to establish and exploit some form of preference order among the potential repairs. We study the problem of computing prioritized repairs over graph databases with data values, using a notion of consistency based on GXPath expressions as integrity constraints. We present several preference criteria based on the standard subset repair semantics, incorporating weights, multisets, and set-based priority levels. We show that it is possible to maintain the same computational complexity as in the case where no preference criterion is available for exploitation. Finally, we explore the complexity of consistent query answering in this setting and obtain tight lower and upper bounds for all the preference criteria introduced., Comment: Appendix

Attribution scores reflect how important the feature values in an input entity are for the output of a machine learning model. One of the most popular attribution scores is the SHAP score, which is an instantiation of the general Shapley value used in coalition game theory. The definition of this score relies on a probability distribution on the entity population. Since the exact distribution is generally unknown, it needs to be assigned subjectively or be estimated from data, which may lead to misleading feature scores. In this paper, we propose a principled framework for reasoning on SHAP scores under unknown entity population distributions. In our framework, we consider an uncertainty region that contains the potential distributions, and the SHAP score of a feature becomes a function defined over this region. We study the basic problems of finding maxima and minima of this function, which allows us to determine tight ranges for the SHAP scores of all features. In particular, we pinpoint the complexity of these problems, and other related ones, showing them to be NP-complete. Finally, we present experiments on a real-world dataset, showing that our framework may contribute to a more robust feature scoring., Comment: In ECAI 2024 proceedings

Recent advancements in immune sequencing and experimental techniques are generating extensive T cell receptor (TCR) repertoire data, enabling the development of models to predict TCR binding specificity. Despite the computational challenges due to the vast diversity of TCRs and epitopes, significant progress has been made. This paper discusses the evolution of the computational models developed for this task, with a focus on machine learning efforts, including the early unsupervised clustering approaches, supervised models, and the more recent applications of Protein Language Models (PLMs). We critically assess the most prominent models in each category, and discuss recurrent challenges, such as the lack of generalization to new epitopes, dataset biases, and biases in the validation design of the models. Furthermore, our paper discusses the transformative role of transformer-based protein models in bioinformatics. These models, pretrained on extensive collections of unlabeled protein sequences, can convert amino acid sequences into vectorized embeddings that capture important biological properties. We discuss recent attempts to leverage PLMs to deliver very competitive performances in TCR-related tasks. Finally, we address the pressing need for improved interpretability in these often opaque models, proposing strategies to amplify their impact in the field.

The ANAIS experiment is intended to search for dark matter annual modulation with ultrapure NaI(Tl) scintillators in order to provide a model independent confirmation or refutation of the long-standing DAMA/LIBRA positive annual modulation signal in the low energy detection rate, using the same target and technique. Other experiments exclude the region of parameters singled out by DAMA/LIBRA. However, these experiments use different target materials, so the comparison of their results depends on the models assumed for the dark matter particle and its distribution in the galactic halo. ANAIS-112, consisting of nine 12.5 kg NaI(Tl) modules produced by Alpha Spectra Inc., disposed in a 3$\times$3 matrix configuration, is taking data smoothly with excellent performance at the Canfranc Underground Laboratory, Spain, since August, 2017. Last published results corresponding to three-year exposure were compatible with the absence of modulation and incompatible with DAMA/LIBRA for a sensitivity above 2.5$\sigma$ C.L. Present status of the experiment and a reanalysis of the first 3 years data using new filtering protocols based on machine-learning techniques are reported. This reanalysis allows to improve the sensitivity previously achieved for the DAMA/LIBRA signal. Updated sensitivity prospects are also presented: with the improved filtering, testing the DAMA/LIBRA signal at 5$\sigma$ will be within reach in 2025., Comment: Contributed to the TAUP2023 Conference, August-September 2023. To be published in Proceeding of Science. arXiv admin note: substantial text overlap with arXiv:2110.10649, arXiv:2209.14113

Next generation mobile networks are poised to transition from monolithic structures owned and operated by single mobile network operators into multi-stakeholder networks where various parties contribute with infrastructure, resources, and services. However, a federation of networks and services brings along a crucial challenge: Guaranteeing secure and trustworthy access control among network entities of different administrative domains. This paper introduces a novel technical concept and a prototype, outlining and implementing a 5G Service-Based Architecture that utilizes Decentralized Identifiers and Verifiable Credentials instead of traditional X.509 certificates and OAuth2.0 access tokens to authenticate and authorize network functions among each other across administrative domains. This decentralized approach to identity and permission management for network functions reduces the risk of single points of failure associated with centralized public key infrastructures. It unifies access control mechanisms and lays the groundwork for lesser complex and more trustful cross-domain key management for highly collaborative network functions in a multi-party Service-Based Architecture of 6G., Comment: This work has been submitted to the IEEE for possible publication

Protein interactions and assembly formation are fundamental to most biological processes. Predicting the assembly structure from constituent proteins -- referred to as the protein docking task -- is thus a crucial step in protein design applications. Most traditional and deep learning methods for docking have focused mainly on binary docking, following either a search-based, regression-based, or generative modeling paradigm. In this paper, we focus on the less-studied multimeric (i.e., two or more proteins) docking problem. We introduce DockGame, a novel game-theoretic framework for docking -- we view protein docking as a cooperative game between proteins, where the final assembly structure(s) constitute stable equilibria w.r.t. the underlying game potential. Since we do not have access to the true potential, we consider two approaches - i) learning a surrogate game potential guided by physics-based energy functions and computing equilibria by simultaneous gradient updates, and ii) sampling from the Gibbs distribution of the true potential by learning a diffusion generative model over the action spaces (rotations and translations) of all proteins. Empirically, on the Docking Benchmark 5.5 (DB5.5) dataset, DockGame has much faster runtimes than traditional docking methods, can generate multiple plausible assembly structures, and achieves comparable performance to existing binary docking baselines, despite solving the harder task of coordinating multiple protein chains., Comment: Under Review

We introduce two new extensions to the beam search algorithm based on conformal predictions (CP) to produce sets of sequences with theoretical coverage guarantees. The first method is very simple and proposes dynamically-sized subsets of beam search results but, unlike typical CP procedures, has an upper bound on the achievable guarantee depending on a post-hoc calibration measure. Our second algorithm introduces the conformal set prediction procedure as part of the decoding process, producing a variable beam width which adapts to the current uncertainty. While more complex, this procedure can achieve coverage guarantees selected a priori. We provide marginal coverage bounds for each method, and evaluate them empirically on a selection of tasks drawing from natural language processing and chemistry., Comment: 11 pages, 4 figures

Conformal regression provides prediction intervals with global coverage guarantees, but often fails to capture local error distributions, leading to non-homogeneous coverage. We address this with a new adaptive method based on rescaling conformal scores with an estimate of local score distribution, inspired by the Jackknife+ method, which enables the use of calibration data in conformal scores without breaking calibration-test exchangeability. Our approach ensures formal global coverage guarantees and is supported by new theoretical results on local coverage, including an a posteriori bound on any calibration score. The strength of our approach lies in achieving local coverage without sacrificing calibration set size, improving the applicability of conformal prediction intervals in various settings. As a result, our method provides prediction intervals that outperform previous methods, particularly in the low-data regime, making it especially relevant for real-world applications such as healthcare and biomedical domains where uncertainty needs to be quantified accurately despite low sample data., Comment: 24 pages, 7 figures

We propose and demonstrate two practical mitigation methods to attenuate the cosmic muon flux, compatible with experiments involving superconducting qubits: shallow underground sites and specific device orientation. Using a specifically-built cosmic muon detector, we identify underground sites, widely present in urban environments, where significant attenuation of cosmic muon flux, up to a factor 35 for 100-meter depths, can be attained. Furthermore, we employ two germanium wafers in an above-ground laboratory, each equipped with a particle sensor, to show how the orientation of the chip with respect to the sky affects the amount and type of energy deposited on the substrate by ionizing radiation. We observe that the horizontal detector sees more counts at lower energy, while the vertical one is impacted by more particles at higher energy. The methods here described proposed ways to directly understand and reduce the effects of cosmic rays on qubits by attenuating the source of this type of decoherence, complementing existing on-chip mitigation strategies. We expect that both on-chip and off-chip methods combined will become ubiquitous in quantum technologies based on superconducting qubit circuits.

In this paper, we provide uniform bounds for convergence rates of the low frequencies of a parametric family of problems for the Laplace operator posed on a rectangular perforated domain of the plane of height $H$. The perforations are periodically placed along the ordinate axis at a distance $O(\epsilon)$ between them, where $\epsilon$ is a parameter that converges towards zero. Another parameter $\eta$, the Floquet-parameter, ranges in the interval $[-\pi, \pi]$. The boundary conditions are quasi-periodicity conditions on the lateral sides of the rectangle and Neumann over the rest. We obtain precise bounds for convergence rates which are uniform on both parameters $\epsilon$ and $\eta$ and strongly depend on $H$. As a model problem associated with a waveguide, one of the main difficulties in our analysis comes near the nodes of the limit dispersion curves.

Diffusion Schr\"odinger bridges (DSB) have recently emerged as a powerful framework for recovering stochastic dynamics via their marginal observations at different time points. Despite numerous successful applications, existing algorithms for solving DSBs have so far failed to utilize the structure of aligned data, which naturally arises in many biological phenomena. In this paper, we propose a novel algorithmic framework that, for the first time, solves DSBs while respecting the data alignment. Our approach hinges on a combination of two decades-old ideas: The classical Schr\"odinger bridge theory and Doob's $h$-transform. Compared to prior methods, our approach leads to a simpler training procedure with lower variance, which we further augment with principled regularization schemes. This ultimately leads to sizeable improvements across experiments on synthetic and real data, including the tasks of predicting conformational changes in proteins and temporal evolution of cellular differentiation processes.

Discrete stochastic processes are widespread in natural systems with many applications across physics, biochemistry, epidemiology, sociology, and finance. While analytic solutions often cannot be derived, existing simulation frameworks can generate stochastic trajectories compatible with the dynamical laws underlying the random phenomena. However, most simulation algorithms assume the system dynamics are memoryless (Markovian assumption), under which assumption, future occurrences only depend on the present state of the system. Mathematically, the Markovian assumption models inter-event times as exponentially distributed variables, which enables the exact simulation of stochastic trajectories using the seminal Gillespie algorithm. Unfortunately, the majority of stochastic systems exhibit properties of memory, an inherently non-Markovian attribute. Non-Markovian systems are notoriously difficult to investigate analytically, and existing numerical methods are computationally costly or only applicable under strong simplifying assumptions, often not compatible with empirical observations. To address these challenges, we have developed the Rejection-based Gillespie algorithm for non-Markovian Reactions (REGIR), a general and scalable framework to simulate non-Markovian stochastic systems with arbitrary inter-event time distributions. REGIR can achieve arbitrary user-defined accuracy while maintaining the same asymptotic computational complexity as the Gillespie algorithm. We illustrate REGIR's modeling capabilities in three important biochemical systems, namely microbial growth dynamics, stem cell differentiation, and RNA transcription. In all three cases, REGIR efficiently models the underlying stochastic processes and demonstrates its utility to accurately investigate complex non-Markovian systems. The algorithm is implemented as a python library REGIR.

Interpretability of deep learning is widely used to evaluate the reliability of medical imaging models and reduce the risks of inaccurate patient recommendations. For models exceeding human performance, e.g. predicting RNA structure from microscopy images, interpretable modelling can be further used to uncover highly non-trivial patterns which are otherwise imperceptible to the human eye. We show that interpretability can reveal connections between the microscopic appearance of cancer tissue and its gene expression profiling. While exhaustive profiling of all genes from the histology images is still challenging, we estimate the expression values of a well-known subset of genes that is indicative of cancer molecular subtype, survival, and treatment response in colorectal cancer. Our approach successfully identifies meaningful information from the image slides, highlighting hotspots of high gene expression. Our method can help characterise how gene expression shapes tissue morphology and this may be beneficial for patient stratification in the pathology unit. The code is available on GitHub., Comment: Github Repo: https://github.com/maragraziani/interpretableWSItoRNAseq

We present an enrichment of the Hateval corpus of hate speech tweets (Basile et. al 2019) aimed to facilitate automated counter-narrative generation. Comparably to previous work (Chung et. al. 2019), manually written counter-narratives are associated to tweets. However, this information alone seems insufficient to obtain satisfactory language models for counter-narrative generation. That is why we have also annotated tweets with argumentative information based on Wagemanns (2016), that we believe can help in building convincing and effective counter-narratives for hate speech against particular groups. We discuss adequacies and difficulties of this annotation process and present several baselines for automatic detection of the annotated elements. Preliminary results show that automatic annotators perform close to human annotators to detect some aspects of argumentation, while others only reach low or moderate level of inter-annotator agreement.