The reference to assumptions in how practitioners use or interact with machine learning (ML) systems is ubiquitous in HCI and responsible ML discourse. However, what remains unclear from prior works is the conceptualization of assumptions and how practitioners identify and handle assumptions throughout their workflows. This leads to confusion about what assumptions are and what needs to be done with them. We use the concept of an argument from Informal Logic, a branch of Philosophy, to offer a new perspective to understand and explicate the confusions surrounding assumptions. Through semi-structured interviews with 22 ML practitioners, we find what contributes most to these confusions is how independently assumptions are constructed, how reactively and reflectively they are handled, and how nebulously they are recorded. Our study brings the peripheral discussion of assumptions in ML to the center and presents recommendations for practitioners to better think about and work with assumptions.
In software development, due to the lack of knowledge or information, time pressure, complex context, and many other factors, various uncertainties emerge during the development process, leading to assumptions scattered in projects. Being unaware of certain assumptions can result in critical problems (e.g., system vulnerability and failures). The prerequisite of analyzing and understanding assumptions in software development is to identify and extract those assumptions with acceptable effort. In this paper, we proposed a tool (i.e., Assumption Miner) to automatically identify and extract assumptions on GitHub projects. To evaluate the applicability of Assumption Miner, we first presented an example of using the tool to mine assumptions from one large and popular deep learning framework project: the TensorFlow project on GitHub. We then conducted an evaluation of the tool. The results show that Assumption Miner can effectively identify and extract assumptions from the repositories on GitHub.
The science of consciousness has been successful over the last decades. Yet, it seems that some of the key questions remain unanswered. Perhaps, as a science of consciousness, we cannot move forward using the same theoretical commitments that brought us here. It might be necessary to revise some assumptions we have made along the way. In this piece, I offer no answers, but I will question some of these fundamental assumptions. We will try to take a fresh look at the classical question about the neural and explanatory correlates of consciousness. A key assumption is that neural correlates are to be found at the level of spiking responses. However, perhaps we should not simply take it for granted that this assumption holds true. Another common assumption is that we are close to understanding the computations underlying consciousness. I will try to show that computations related to consciousness might be far more complex than our current theories envision. There is little reason to think that consciousness is an abstract computation, as traditionally believed. Furthermore, I will try to demonstrate that consciousness research could benefit from investigating internal changes of consci
Over the past few decades, we have seen a proliferation of advanced cryptographic primitives with lossy or homomorphic properties built from various assumptions such as Quadratic Residuosity, Decisional Diffie-Hellman, and Learning with Errors. These primitives imply hard problems in the complexity class $SZK$ (statistical zero-knowledge); as a consequence, they can only be based on assumptions that are broken in $BPP^{SZK}$. This poses a barrier for building advanced primitives from code-based assumptions, as the only known such assumption is Learning Parity with Noise (LPN) with an extremely low noise rate $\frac{\log^2 n}{n}$, which is broken in quasi-polynomial time. In this work, we propose a new code-based assumption: Dense-Sparse LPN, that falls in the complexity class $BPP^{SZK}$ and is conjectured to be secure against subexponential time adversaries. Our assumption is a variant of LPN that is inspired by McEliece's cryptosystem and random $k\mbox{-}$XOR in average-case complexity. We leverage our assumption to build lossy trapdoor functions (Peikert-Waters STOC 08). This gives the first post-quantum alternative to the lattice-based construction in the original paper. Lossy
Nonconvex optimization is central to modern machine learning, but the general framework of nonconvex optimization yields weak convergence guarantees that are too pessimistic compared to practice. On the other hand, while convexity enables efficient optimization, it is of limited applicability to many practical problems. To bridge this gap and better understand the practical success of optimization algorithms in nonconvex settings, we introduce a novel unified parametric assumption. Our assumption is general enough to encompass a broad class of nonconvex functions while also being specific enough to enable the derivation of a unified convergence theorem for gradient-based methods. Notably, by tuning the parameters of our assumption, we demonstrate its versatility in recovering several existing function classes as special cases and in identifying functions amenable to efficient optimization. We derive our convergence theorem for both deterministic and stochastic optimization, and conduct experiments to verify that our assumption can hold practically over optimization trajectories.
The independence assumption is a useful tool to increase the tractability of one's modelling framework. However, this assumption does not match reality; failing to take dependencies into account can cause models to fail dramatically. The field of multi-axis graphical modelling (also called multi-way modelling, Kronecker-separable modelling) has seen growth over the past decade, but these models require that the data have zero mean. In the multi-axis case, inference is typically done in the single sample scenario, making mean inference impossible. In this paper, we demonstrate how the zero-mean assumption can cause egregious modelling errors, as well as propose a relaxation to the zero-mean assumption that allows the avoidance of such errors. Specifically, we propose the "Kronecker-sum-structured mean" assumption, which leads to models with nonconvex-but-unimodal log-likelihoods that can be solved efficiently with coordinate descent.
We revisit a classical assumption for analyzing stochastic gradient algorithms where the squared norm of the stochastic subgradient (or the variance for smooth problems) is allowed to grow as fast as the squared norm of the optimization variable. We contextualize this assumption in view of its inception in the 1960s, its seemingly independent appearance in the recent literature, its relationship to weakest-known variance assumptions for analyzing stochastic gradient algorithms, and its relevance in deterministic problems for non-Lipschitz nonsmooth convex optimization. We build on and extend a connection recently made between this assumption and the Halpern iteration. For convex nonsmooth, and potentially stochastic, optimization, we analyze horizon-free, anytime algorithms with last-iterate rates. For problems beyond simple constrained optimization, such as convex problems with functional constraints or regularized convex-concave min-max problems, we obtain rates for optimality measures that do not require boundedness of the feasible set.
Conformal predictors provide set or functional predictions that are valid under the assumption of randomness, i.e., under the assumption of independent and identically distributed data. The question asked in this paper is whether there are predictors that are valid in the same sense under the assumption of randomness and that are more efficient than conformal predictors. The answer is that the class of conformal predictors is universal in that only limited gains in predictive efficiency are possible. The previous work in this area has relied on the algorithmic theory of randomness and so involved unspecified constants, whereas this paper's results are much more practical. They are also shown to be optimal in some respects.
The ubiquitous independence assumption among symbolic concepts in neurosymbolic (NeSy) predictors is a convenient simplification: NeSy predictors use it to speed up probabilistic reasoning. Recent works like van Krieken et al. (2024) and Marconato et al. (2024) argued that the independence assumption can hinder learning of NeSy predictors and, more crucially, prevent them from correctly modelling uncertainty. There is, however, scepticism in the NeSy community around the scenarios in which the independence assumption actually limits NeSy systems (Faronius and Dos Martires, 2025). In this work, we settle this question by formally showing that assuming independence among symbolic concepts entails that a model can never represent uncertainty over certain concept combinations. Thus, the model fails to be aware of reasoning shortcuts, i.e., the pathological behaviour of NeSy predictors that predict correct downstream tasks but for the wrong reasons.
Naturally occurring information-seeking questions often contain questionable assumptions -- assumptions that are false or unverifiable. Questions containing questionable assumptions are challenging because they require a distinct answer strategy that deviates from typical answers for information-seeking questions. For instance, the question "When did Marie Curie discover Uranium?" cannot be answered as a typical "when" question without addressing the false assumption "Marie Curie discovered Uranium". In this work, we propose (QA)$^2$ (Question Answering with Questionable Assumptions), an open-domain evaluation dataset consisting of naturally occurring search engine queries that may or may not contain questionable assumptions. To be successful on (QA)$^2$, systems must be able to detect questionable assumptions and also be able to produce adequate responses for both typical information-seeking questions and ones with questionable assumptions. Through human rater acceptability on end-to-end QA with (QA)$^2$, we find that current models do struggle with handling questionable assumptions, leaving substantial headroom for progress.
Political scientists are increasingly interested in assessing causal mechanisms, or determining not just if a causal effect exists but also why it occurs. Even so, many researchers avoid formal causal mediation analyses due to their stringent assumptions, instead opting to explore causal mechanisms through what we call intermediate outcome tests. These tests estimate the effect of the treatment on one or more mediators and view such effects as suggestive evidence of a causal mechanism. In this paper, we use nonparametric bounding analysis to show that, without further assumptions, these tests can neither establish nor rule out the existence of a causal mechanism. To use intermediate outcome tests as a falsification test of causal mechanisms, researchers must make a very strong but rarely discussed monotonicity assumption. We develop a way to assess the plausibility of this monotonicity assumption and estimate our bounds for two recent experiments that use these tests.
Sensitivity analysis informs causal inference by assessing the sensitivity of conclusions to departures from assumptions. The consistency assumption states that there are no hidden versions of treatment and that the outcome arising naturally equals the outcome arising from intervention. When reasoning about the possibility of consistency violations, it can be helpful to distinguish between covariates and versions of treatment. In the context of surgery, for example, genomic variables are covariates and the skill of a particular surgeon is a version of treatment. There may be hidden versions of treatment, and this paper addresses that concern with a new kind of sensitivity analysis. Whereas many methods for sensitivity analysis are focused on confounding by unmeasured covariates, the methodology of this paper is focused on confounding by hidden versions of treatment. In this paper, new mathematical notation is introduced to support the novel method, and example applications are described.
Event Causality Identification (ECI) aims at determining whether there is a causal relation between two event mentions. Conventional prompt learning designs a prompt template to first predict an answer word and then maps it to the final decision. Unlike conventional prompts, we argue that predicting an answer word may not be a necessary prerequisite for the ECI task. Instead, we can first make a deterministic assumption on the existence of causal relation between two events and then evaluate its rationality to either accept or reject the assumption. The design motivation is to try the most utilization of the encyclopedia-like knowledge embedded in a pre-trained language model. In light of such considerations, we propose a deterministic assumption prompt learning model, called DAPrompt, for the ECI task. In particular, we design a simple deterministic assumption template concatenating with the input event pair, which includes two masks as predicted events' tokens. We use the probabilities of predicted events to evaluate the assumption rationality for the final event causality decision. Experiments on the EventStoryLine corpus and Causal-TimeBank corpus validate our design objective
There are many kinds of exogeneity assumptions. How should researchers choose among them? When exogeneity is imposed on an unobservable like a potential outcome, we argue that the form of exogeneity should be chosen based on the kind of selection on unobservables it allows. Consequently, researchers can assess the plausibility of any exogeneity assumption by studying the distributions of treatment given the unobservables that are consistent with that assumption. We use this approach to study two common exogeneity assumptions: quantile and mean independence. We show that both assumptions require a kind of non-monotonic relationship between treatment and the potential outcomes. We discuss how to assess the plausibility of this kind of treatment selection. We also show how to define a new and weaker version of quantile independence that allows for monotonic treatment selection. We then show the implications of the choice of exogeneity assumption for identification. We apply these results in an empirical illustration of the effect of child soldiering on wages.
We show that an integral assumption in DGLV radiative energy loss - the large formation time assumption - is violated at high-$p_T$ for phenomenologically relevant parameters. We further investigate the phenomenological impact of placing a new kinematic bound on the radiated gluon transverse momentum, which ensures that there are no contributions to the energy loss from regions of parameter space that violate the large formation time assumption. We find that this places a large sensitivity on the exact kinematic cutoff used, similar to the known collinear cutoff sensitivity, indicating the theoretical need for a rederivation of DGLV radiative energy with the large formation time assumption relaxed in order to make rigorous predictions. We additionally find that this large formation time cutoff dramatically reduces the size of a short pathlength correction to the DGLV radiative energy loss, which is of phenomenological interest in predicting suppression in small $p +A$ systems. We compute the phenomenological predictions utilizing this large formation time cutoff in both $p+A$ and $A+A$ collisions at the LHC, in a convolved radiative and elastic energy loss model.
State-of-the-art neurosymbolic learning systems use probabilistic reasoning to guide neural networks towards predictions that conform to logical constraints over symbols. Many such systems assume that the probabilities of the considered symbols are conditionally independent given the input to simplify learning and reasoning. We study and criticise this assumption, highlighting how it can hinder optimisation and prevent uncertainty quantification. We prove that loss functions bias conditionally independent neural networks to become overconfident in their predictions. As a result, they are unable to represent uncertainty over multiple valid options. Furthermore, we prove that these loss functions are difficult to optimise: they are non-convex, and their minima are usually highly disconnected. Our theoretical analysis gives the foundation for replacing the conditional independence assumption and designing more expressive neurosymbolic probabilistic models.
This paper studies the identification, estimation, and hypothesis testing problem in complete and incomplete economic models with testable assumptions. Testable assumptions ($A$) give strong and interpretable empirical content to the models but they also carry the possibility that some distribution of observed outcomes may reject these assumptions. A natural way to avoid this is to find a set of relaxed assumptions ($\tilde{A}$) that cannot be rejected by any distribution of observed outcome and the identified set of the parameter of interest is not changed when the original assumption is not rejected. The main contribution of this paper is to characterize the properties of such a relaxed assumption $\tilde{A}$ using a generalized definition of refutability and confirmability. I also propose a general method to construct such $\tilde{A}$. A general estimation and inference procedure is proposed and can be applied to most incomplete economic models. I apply my methodology to the instrument monotonicity assumption in Local Average Treatment Effect (LATE) estimation and to the sector selection assumption in a binary outcome Roy model of employment sector choice. In the LATE applicatio
Causal mediation analysis is a useful tool for epidemiological research, but it has been criticized for relying on a "cross-world" independence assumption that is empirically difficult to verify and problematic to justify based on background knowledge. In the present article we aim to assist the applied researcher in understanding this assumption. Synthesizing what is known about the cross-world independence assumption, we discuss the relationship between assumptions for causal mediation analyses, causal models, and non-parametric identification of natural direct and indirect effects. In particular we give a practical example of an applied setting where the cross-world independence assumption is violated even without any post-treatment confounding. Further, we review possible alternatives to the cross-world independence assumption, including the use of computation of bounds that avoid the assumption altogether. Finally, we carry out a numerical study in which the cross-world independence assumption is violated to assess the ensuing bias in estimating natural direct and indirect effects. We conclude with recommendations for carrying out causal mediation analyses.
A representative researcher has repeated opportunities for empirical research. To process findings, she must impose an "identifying assumption." She conducts research when the assumption is sufficiently plausible (taking into account both current beliefs and the quality of the opportunity), and updates beliefs as if the assumption were perfectly valid. We study the dynamics of this learning process. While the rate of research cannot always increase over time, research slowdown is possible. We characterize environments in which the rate is constant. Long-run beliefs can exhibit history-dependence and "false certitude." We apply the model to stylized examples of empirical methodologies: experiments, various causal-inference techniques, and "calibration."
The synthesis problem asks to construct a reactive finite-state system from an $ω$-regular specification. Initial specifications are often unrealizable, which means that there is no system that implements the specification. A common reason for unrealizability is that assumptions on the environment of the system are incomplete. We study the problem of correcting an unrealizable specification $φ$ by computing an environment assumption $ψ$ such that the new specification $ψ\toφ$ is realizable. Our aim is to construct an assumption $ψ$ that constrains only the environment and is as weak as possible. We present a two-step algorithm for computing assumptions. The algorithm operates on the game graph that is used to answer the realizability question. First, we compute a safety assumption that removes a minimal set of environment edges from the graph. Second, we compute a liveness assumption that puts fairness conditions on some of the remaining environment edges. We show that the problem of finding a minimal set of fair edges is computationally hard, and we use probabilistic games to compute a locally minimal fairness assumption.