Deep Multimodal Data Fusion

ACM Computing Surveys, Journal Year: 2024, Volume and Issue: 56(9), P. 1 - 36

Published: Feb. 24, 2024

Multimodal Artificial Intelligence (Multimodal AI), in general, involves various types of data (e.g., images, texts, or collected from different sensors), feature engineering extraction, combination/fusion), and decision-making majority vote). As architectures become more sophisticated, multimodal neural networks can integrate fusion, processes into one single model. The boundaries between those are increasingly blurred. conventional fusion taxonomy early/late fusion), based on which the occurs in, is no longer suitable for modern deep learning era. Therefore, main-stream techniques used, we propose a new fine-grained grouping state-of-the-art (SOTA) models five classes: Encoder-Decoder methods, Attention Mechanism Graph Neural Network Generative other Constraint-based methods. Most existing surveys only focused specific task with combination two modalities. Unlike those, this survey covers broader modalities, including Vision + Language videos, texts), Sensors LiDAR), so on, their corresponding tasks video captioning, object detection). Moreover, comparison among these methods provided, as well challenges future directions area.

Language: Английский

---

Adolescent-to-adult gains in cognitive flexibility are adaptively supported by reward sensitivity, exploration, and neural variability

Current Opinion in Behavioral Sciences, Journal Year: 2024, Volume and Issue: 58, P. 101399 - 101399

Published: May 10, 2024

Language: Английский

---

The “silent” noise: moving forward from bias to noise in football referees’ decision-making

International Journal of Sport and Exercise Psychology, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 26

Published: Jan. 21, 2025

In football, referees are expected to deliver consistent and unbiased judgments, grounded in professional knowledge expertise. However, much of the research on referees' decision-making has traditionally focused concept "bias" their judgments. This review shifts attention noise – defined as "undesirable variability judgments same problem" (Kahneman, D., Sibony, O., & Sunstein, C. R. [2021]. Noise: A flaw human judgment (p. 40). Little, Brown Spark). Noise reflects inconsistency responses similar match situations, resulting diverse decisions for comparable infringements. The article is structured into five key sections. First, we introduce judgment. Second, explore issue context football refereeing, offering examples relevant data. Particular emphasis placed foul decisions, incorporating both raw data findings from existing literature. detailed framework presented, outlining components, sources, detection methods, strategies reducing refereeing. third section, compare bias officiating, examining potential mechanisms underlying each. fourth section considers errors refereeing discusses costs associated with implementing such measures. Finally, argue why stakeholders officiating should expand focus beyond address implications decisions.

Language: Английский

---

Noise in Cognition: Bug or Feature?

Perspectives on Psychological Science, Journal Year: 2025, Volume and Issue: unknown

Published: March 4, 2025

Noise in behavior is often considered a nuisance: Although the mind aims for best possible action, it let down by unreliability sensory and response systems. Researchers represent noise as additive, Gaussian, independent. Yet careful look at behavioral reveals rich structure that defies easy explanation. First, both perceptual preferential judgments may potentially play only minor roles, with most arising cognitive computations. Second, functional form of non-Gaussian nonindependent, distribution being better characterized heavy-tailed having substantial long-range autocorrelations. It this results from brains are, some reason, bedeviled fundamental design flaw, albeit one intriguingly distinctive characteristics. Alternatively, might not be bug but feature. Specifically, we propose brain approximates probabilistic inference local sampling algorithm, using randomness to drive its exploration alternative hypotheses. Reframing cognition way explains leads surprising conclusion symptom malfunction plays central role underpinning human intelligence.

Language: Английский

---

Compulsivity is linked to suboptimal choice variability but unaltered reinforcement learning under uncertainty

Nature Mental Health, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 6, 2025

Language: Английский

---

Rational arbitration between statistics and rules in human sequence processing

Nature Human Behaviour, Journal Year: 2022, Volume and Issue: 6(8), P. 1087 - 1103

Published: May 2, 2022

Language: Английский

---

Impaired value-based decision-making in Parkinson’s disease apathy

Brain, Journal Year: 2024, Volume and Issue: 147(4), P. 1362 - 1376

Published: Feb. 2, 2024

Abstract Apathy is a common and disabling complication of Parkinson’s disease characterized by reduced goal-directed behaviour. Several studies have reported dysfunction within prefrontal cortical regions projections from brainstem nuclei whose neuromodulators include dopamine, serotonin noradrenaline. Work in animal human neuroscience confirmed contributions these on aspects motivated decision-making. Specifically, overlapping to encoding the value decisions, influence whether explore alternative courses action or persist an existing strategy achieve rewarding goal. Building upon this work, we hypothesized that apathy should be associated with impairment value-based learning. Using four-armed restless bandit reinforcement learning task, studied decision-making 75 volunteers; 53 patients disease, without clinical apathy, 22 age-matched healthy control subjects. Patients exhibited impaired ability choose highest bandit. Task performance predicted individual patient’s severity measured using Lille Rating Scale (R = −0.46, P < 0.001). Computational modelling choices group made decisions were indifferent learnt options, consistent previous reports reward insensitivity. Further analysis demonstrated shift away exploiting option reduction perseveration, which also correlated scores −0.5, We went acquire functional MRI 59 19 20 controls performing Restless Bandit Task. Analysis signal at point feedback diminished ventromedial cortex was more marked but not predictive their severity. model-based categorization choice type, lower bandits activated similar degree controls. In contrast, significantly increased activation across distributed thalamo-cortical network. Enhanced activity thalamus both patient groups connectivity dorsal anterior cingulate insula. Given task no different subjects, interpret recruitment network as possible compensatory mechanism, compensates against symptomatic manifestation disease.

Language: Английский

---

Anticipatory Anxiety and Wishful Thinking

American Economic Review, Journal Year: 2024, Volume and Issue: 114(4), P. 926 - 960

Published: March 28, 2024

Across five experiments (N = 1,714), we test whether people engage in wishful thinking to alleviate anxiety about adverse future outcomes. Participants perform pattern recognition tasks which some patterns may result an electric shock or a monetary loss. Diagnostic of thinking, participants are less likely correctly identify that associated with Wishful is more pronounced under ambiguous signals and only reduced by higher accuracy incentives when participants’ cognitive effort reduces ambiguity. disappears the domain gains, indicating negative emotions important drivers phenomenon. (JEL C91, D12, D83, D91)

Language: Английский

---

Insensitive Investors

The Journal of Finance, Journal Year: 2024, Volume and Issue: 79(4), P. 2473 - 2503

Published: June 18, 2024

ABSTRACT We experimentally study the transmission of subjective expectations into actions. Subjects in our experiment report valuations that are far too insensitive to their expectations, relative prediction from a frictionless model. propose insensitivity is driven by noisy cognitive process prevents subjects precisely computing asset valuations. The empirical link between and actions becomes stronger as approach rational expectations. Our results highlight importance incorporating weak belief‐based pricing models. Finally, we discuss how noise can provide microfoundation for inelastic demand stock market.

Language: Английский

---

Dynamic noise estimation: A generalized method for modeling noise fluctuations in decision-making

Journal of Mathematical Psychology, Journal Year: 2024, Volume and Issue: 119, P. 102842 - 102842

Published: Feb. 28, 2024

Computational cognitive modeling is an important tool for understanding the processes supporting human and animal decision-making. Choice data in decision-making tasks are inherently noisy, separating noise from signal can improve quality of computational modeling. Common approaches to model decision often assume constant levels or exploration throughout learning (e.g., ϵ-softmax policy). However, this assumption not guaranteed hold – example, a subject might disengage lapse into inattentive phase series trials middle otherwise low-noise performance. Here, we introduce new, computationally inexpensive method dynamically estimate fluctuations choice behavior, under that agent transition between two discrete latent states fully engaged random). Using simulations, show instead statically substantially fit parameter estimation, especially presence long periods noisy such as prolonged lapses attention. We further demonstrate empirical benefits dynamic estimation at individual group by validating it on four published datasets featuring diverse populations, tasks, models. Based theoretical evaluation reported current work, expect will many paradigms over static currently used literature, while keeping additional complexity assumptions minimal.

Language: Английский

---