Beyond Robots - embodying the fantasy and piling up the imagination.

Theory of Mind is often characterized as the ability to recognize desires, beliefs, and intentions of others. In this position paper, I look at the literature on modeling Theory of Mind in machines and find that, to date, intention is not usually a focus. I define what I mean by intention---choice with commitment---following prior work. Intention has a long history of research in some communities, and I offer one theoretical framework for modeling intention as a starting point. I take inspiration from how children learn intention through joint attention with others and how that leads to Theory of Mind. I argue that though models of machine Theory of Mind need not follow the same learning progression as children, intention is an aspect of Theory of Mind that should be more explicit.

AI bias in Human-Robot Interaction: An evaluation of the Risk in Gender Biased Robots

Machine-in-the-loop writing aims to build models that assist humans to accomplish their writing tasks more effectively. Prior work has 
found that providing users a machine-written draft or sentence-level continuations has limited success since the generated text tends to deviate from users' intention.  To allow the user to retain control over the content, we train a rewriting model that, when prompted, modifies specified spans of text within the user's original draft to introduce descriptive and figurative elements in the text.
We evaluate the model on its ability to collaborate with humans on the task of creative image captioning. On a user study through Amazon Mechanical Turk, our model is rated to be more helpful by users than a baseline infilling language model. In addition, third-party evaluation shows that users write more descriptive and figurative captions when collaborating with our model compared to completing the task alone.
However, the improvement is not uniform across user groups: the model is more helpful to skilled users, which risks widening the gap between skilled and novice users, highlighting a need for careful, user-centric evaluation of interactive systems.

Machine-in-the-Loop Rewriting for Creative Image Captioning

Twitter is one of the most popular social networks attracting millions of users, while a considerable proportion of online discourse is captured. It provides a simple usage framework with short messages and an efficient application programming interface (API) enabling the research community to study and analyze several aspects of this social network. However, the Twitter usage simplicity can lead to malicious handling by various bots. The malicious handling phenomenon expands in online discourse, especially during the electoral periods, where except the legitimate bots used for dissemination and communication purposes, the goal is to manipulate the public opinion and the electorate towards a certain direction, specific ideology, or political party.

This paper focuses on the design of a novel system for identifying Twitter bots based on labeled Twitter data. To this end, a supervised machine learning (ML) framework is adopted using an Extreme Gradient Boosting (XGBoost) algorithm, where the hyper-parameters are tuned via cross-validation. Our study also deploys Shapley Additive Explanations (SHAP) for explaining the ML model predictions by calculating feature importance, using the game theoretic-based Shapley values. Experimental evaluation on distinct Twitter datasets demonstrate the superiority of our approach, in terms of bot detection accuracy, when compared against a recent state-of-the-art Twitter bot detection method.

Identification of Twitter Bots Based on an Explainable Machine Learning Framework: The US 2020 Elections Case Study

Bias in news reporting can lead to tribalism and division on important issues. Scalable and reliable measurement of such biases is an important first step in addressing them. In this work, based on the intuition that media bias is captured by the tone and word choices in articles, we propose a framework for modeling the relative bias of media outlets through masked token prediction via large-scale pretrained masked language models fine-tuned on articles form news outlets. Through experiments on five diverse and politically polarized topics we show that our framework can capture media bias towards these topics with high reliability. Additionally, our experiments show that our framework is general, in that language models fine-tuned on one topic can be applied to other topics with little drop in performance.

Measuring Media Bias via Masked Language Modeling

The COVID-19 (Coronavirus Disease 2019), caused by the SARS-CoV-2 virus, can cause severe pneumonia. The chest X-ray (CXR) is the preliminary radiographic exam to investigate pneumonia and other lung abnormalities. However, the identification of opacities in the CXR is challenging due to its overall low quality. In the literature, several works have proposed automated approaches to improve the quality of CXR images in general. However, as far as we know, few studies explored them in the context of COVID-19 identification in a thorough manner. In this work, we applied a series of digital image processing techniques, focusing on contrast enhancement. The impact of each method was evaluated using traditional digital image quality metrics, specifically peak signal-to-noise ratio (PSNR) and universal image quality (UQI), and a deep learning model for COVID-19 detection. We also investigated which regions were most frequently used by the model after each transformation using an Explainable Artificial Intelligence (XAI) approach called LIME. The rationale is to evaluate if any transformation improves the lung region used by the recognition model. The enhancement strategies that presented the overall best results considering PSNR and UQI were Gamma Correction (GC) and Adaptive Gamma Correction With Weighting Distribution (AGCWD). However, no method improved the identification of COVID-19 in a significant matter.

Chest X-ray image enhancement: a COVID-19 identification case study

Anthrax is a fatal disease caused by bacteria called bacillus anthracis. The bacteria can be used as a bioterror agent due to its high toxicity and resistance to the external environment of its spore. Dipicolinic acid (DPA) is released when the spore germinates; therefore, sensitively detecting DPA as a biomarker of the spores is focused on by many researchers [1]. Among them, nano-sensors including trivalent lanthanide ions have been used to detect DPA sensitively and promptly because the trivalent lanthanide ion such as europium (Eu3+) and terbium (Tb3+) shows unique fluorescent properties originated from their energy levels, and the intensity can be enhanced by chelating with DPA called antenna effect [2]. However, nanomaterials such as polymers or inorganic nanoparticles containing fluorescent substances usually quench the fluorescence by absorbing the excitation and emission light concurrently [3]. Therefore it is necessary to separate the fluorescent signal from the nanomaterials to improve its sensitivity. Herein, we develop terbium ion-coordinated iron oxide (Fe3O4-Tb) nanoparticles and detect DPA via measuring fluorescent signals from the Tb-DPA complexes [4]. The Fe3O4-Tb nanoparticles have a magnetite (Fe3O4) core and the surface is enveloped by Tb3+ that is coordinated with polyacrylate. Because of the small crystallite size of the Fe3O4-Tb nanoparticle, it has a trivial coercivity; therefore, Fe3O4-Tb nanoparticles do not agglomerate each other and reacts DPA with high reproducibility [5]. When the Fe3O4-Tb nanoparticles contact with DPA, Tb3+ is disassociated from the nanoparticle surface and chelated with DPA. The formed Tb-DPA complexes have a green fluorescence under certain ultraviolet (UV) wavelengths. We separate the iron oxide core via a permanent magnet to prevent quenching of the nanoparticles, then the clear signal shows a lower detection limit (5.1 nM) than the infectious dosage of anthrax (60 μM) [6]. The detecting method also has a high selectivity when compared with other aromatic carboxylic acids, which have a molecular structure similar to DPA.

![](https://assets.underline.io/uploads/markdown_image/1/image/14a2f16e4463e6542107f235846a8f38.jpg)

Fig.1. Detection mechanism for DPA with Fe3O4-Tb nanoparticles and the fluorescent intensity dependent on the concentration of DPA.

Fluorescent detection of dipicolinic acid as a biomarker for anthrax using Tb ion-coordinated Fe3O4 nanoparticles

High-entropy alloys (HEAs), whose new alloy design concept with multiple principal elements, attracted a lot of attention for structural application and functional aspects. Since their first report in 2004, their compositions have developed from first generation with quinary equiatomic single-phase to second-generation multi-phase structure with non-equiatomic compositions. For magnetocaloric HEAs, their limitations have been recently found to be overcome by using non-equiatomic HEA compositions: GdTbCoAlFe (with Curie temperatures up to 108 K) overcomes the low temperature limit of the rare-earth-containing HEAs (typically saturates around &lt; 60 K) [1,2] and the first-order magnetostructural phase transition in FeMnNiGeSi surpasses the smeared out magnetocaloric effect (MCE) of rare-earth-free HEAs [3,4]. In this work, we further enhance the MCE of non-equiatomic Gd34.9Tb19.4Co19Al23.3Fe3 HEA through microstructural control, in which the alloy microwires are annealed using the current density annealing technique. Induced nanocrystals in the amorphous matrix observed from transmission electron microscopy (TEM) are found to increase and grow with the current density magnitude (Fig. 1). This leads to an increase in the working temperature span of the HEA microwires and at the same time offers a relative cooling power (RCP) comparable to many reported amorphous MCE alloys, both conventional alloys and HEAs (Fig. 2). Our findings pave a pathway for further optimizing MCE through appropriate processing methods on top of careful alloy design selection.

![](https://assets.underline.io/uploads/markdown_image/1/image/fce97f9433fab8f05b035a2bf2ea02e2.jpg)

Fig. 1 TEM results of the (a) as-cast and annealed Gd34.9Tb19.4Co19Al23.3Fe3 microwires using current density values of (b) 50 and (c) 100 x 106 A m-2.

![](https://assets.underline.io/uploads/markdown_image/1/image/a7e0bd8f7ddc93b92bbc79c9201b5616.jpg)

Fig. 2 Literature comparison of RCP as a function of full-width at half maximum (FWHM) for studied microwires (star symbols) versus amorphous conventional alloys (triangles) and HEAs (squares).

Enhanced Magnetocaloric Response of High-Entropy Alloy Microwires due to Induced Nanocrystals by Current Annealing

In superparamagnetic tunnel junctions (SMTJ), the energy barriers that separate the parallel and the antiparallel magnetization configurations are sufficiently small that the thermal noise causes stochastic magnetization fluctuations between the two states. Through magnetoresistance effects, those stochastic fluctuations give rise to telegraphic noise signals with two voltage levels (see Fig.1) [1, 2]. The relative average time spent in each of the two configurations can be controlled by an applied dc current via the spin-transfer torque. Due to their inherent stochastic nature, SMTJs are promising candidates for truly random number generator applications [3, 4], brain-inspired and probabilistic computing [5-10]. However, an external bias magnetic field was always required to fully operate the SMTJ’s with current, which represents a drawback to their integration with the required microelectronic circuits. In this work, we report for the first time, SMTJs that are fully controlled by DC current at zero applied magnetic field using perpendicular magnetic tunnel junctions (pMTJ). Stable pMTJs are used in densely magnetic memories [11], nevertheless their thermal stability can be strongly reduced by tuning their free layer thickness, leading to a superparamagnetic state. Here, we use pMTJ’s with a free layer thickness of t=1.6 nm, and with diameters ranging from 150 nm to 80 nm. By applying a DC current in one direction (sign), without any external bias field, we observe stochastic fluctuations in the resistance for dc currents between -0.3 mA and -0.27 mA. As shown in Fig. 2, the average probability in being in high and low resistance state is well controlled by the applied current. In order to obtain good statistics on the extracted probability, we record to stochastic transitions. We extracted the dwell time distribution, which follows an exponential distribution with a mean dwell time that is in agreement with the modified Néel-Brown model. The authors acknowledge partial funding from ERC MAGICAL No 669204, and the Nanoscience Foundation (Grenoble, France).

![](https://assets.underline.io/uploads/markdown_image/1/image/032d011551d8698e3dc1b74244af9c81.jpg)

Figure 1 : Experimental voltage time trace of the SMTJ.

![](https://assets.underline.io/uploads/markdown_image/1/image/f8c0fe8daabe77b0c738c5d388eadce3.jpg)

Figure 2 : The probability of occupancy of the high resistance state versus current.

Current controlled perpendicular superparamagnetic tunnel junctions operating at zero applied magnetic field

Early detection of disease biomarkers, toxins, and infectious microorganisms is a major goal in modern health care, environmental control, and food safety to help contain and reverse illness or prevent contaminants from entering the food chain. To achieve this, materials scientists develop biosensing platforms whose requirements may differ depending on the application. When biosensing is a tool for screening or point-of-use applications (e.g., at the patient's bedside, in the industrial plant, or harsh environments), it must be fast, low-cost, and easy to use and transport. Lateral flow immunoassays (LFI) meet these requirements. They are paper-based tests, whose best-known example is the home pregnancy test, developed to detect high levels of human chorionic gonadotropin in urine. Thanks to advances in immunology, the method has been extended with great success to a large variety of targets, the most recent the rapid diagnostic tests for SARS-CoV2 [1]. There are some limitations for LFIA expansion, mainly their sensitivity and detection limit, which are not sufficient for some applications. Magnetic nanoparticles and magnetic detectors can be an excellent way to overcome such constraints. LFI (Fig. 1) uses a strip of nitrocellulose along which a liquid sample (urine, saliva, blood, serum, or plasma) can flow by capillary action. The target molecule is selectively captured on the strip by an immunological reaction and tagged by a reporter particle or enzyme that makes it detectable. For this purpose, plasmonic or fluorescent nanoparticles, visible with the naked eye, provide the positive/negative response. Although this is good enough for pregnancy, diagnoses frequently demand quantifying biomarkers, cells, bacteria, viruses, genes, or toxins. Quantification of color reporters involves image analysis, reflectance, or fluorescence measurements. However, these methods are sensitive to ambient light, humidity, and staining of the paper strip, which cause difficulties in calibration and reproducibility, especially in samples with a complex matrix [2]. To improve upon this, ferrite magnetic nanoparticles are now being used as alternative LFI reporters [3]. Although their bio-functionalization can be more challenging than for plasmonic particles, their advantages are clear for many applications. First, iron oxide nanoparticles provide a brownish color that is detectable by the naked eye or on a smartphone camera image. Magnetism can be used for pre-concentration or isolation of the target analyte from the sample matrix, and the magnetic reporters can be relocated closer to the visible membrane surface to increase their visibility. Second, magnetic nanoparticles produce a magnetic perturbation that a magnetic sensor can detect without optical interference [2]. Additionally, magnetic sensors report particles not only from the paper's surface but from the entire cross-section of the strip [4]. Finally, the magnetic signal from the nanotags does not decrease significantly with time. We review the principles and design of magnetic LFI and detail the requirements for magnetic nanoparticles to be used as reporters. We address the current methods used to read their signals that do not sacrifice the simplicity and low cost of the paper-based method. As proof of how magnetic LFI can be a useful analytical tool for biomedicine and food safety applications, we give examples of pneumonia biomarker detection (Fig. 2) and toxin quantification in foods and beverages [2], [4]. We acknowledge support from the Spanish Ministry of Science and Innovation (EIN2020-112354), the Principality of Asturias (IDI/2018/000185), and the Instituto Universitario de Tecnología Industrial de Asturias (SV-20-GIJÓN-1-22).

![](https://assets.underline.io/uploads/markdown_image/1/image/55e5c243e8177961bca7912e0981d64c.jpg)

Magnetic sensor reading out of a lateral flow strip for pneumolysin detection. The calibration yields a limit of detection of 0.2 ng of protein per mL of sample. Inset: Scheme (left) and image (right) of a magnetic lateral flow immunoassay.

Magnetic Techniques for Rapid Diagnostic Testing for Health Care and Environmental Monitoring

The importance of diversity for excellence in science

Machine-in-the-loop writing aims to enable humans to collaborate with models to effectively complete their writing tasks. Prior work in the creative domain has found that providing humans with a machine written draft or sentence level continuations has limited success as the generated text tends to deviate from the human's intentions. We train a rewriting model that, when prompted, modifies targeted spans of text within the user's original draft, enabling the human to retain control over the content while still taking advantage of the strengths of text generation models to introduce descriptive and figurative elements locally in the text. We evaluate the model on it's ability to collaborate with humans on the task of creative image captioning through a user study on Amazon Mechanical Turk. Users report that the model is helpful and third-party evaluation shows that users write more descriptive and figurative captions on average in the collaborative setting compared to a baseline of the human completing the task alone.

Underline digital video library

Beyond Robots - embodying the fantasy and piling up the imagination.

AI bias in Human-Robot Interaction: An evaluation of the Risk in Gender Biased Robots

Machine-in-the-Loop Rewriting for Creative Image Captioning

Identification of Twitter Bots Based on an Explainable Machine Learning Framework: The US 2020 Elections Case Study

Measuring Media Bias via Masked Language Modeling

Chest X-ray image enhancement: a COVID-19 identification case study

Fluorescent detection of dipicolinic acid as a biomarker for anthrax using Tb ion-coordinated Fe3O4 nanoparticles

Enhanced Magnetocaloric Response of High-Entropy Alloy Microwires due to Induced Nanocrystals by Current Annealing

Current controlled perpendicular superparamagnetic tunnel junctions operating at zero applied magnetic field

Magnetic Techniques for Rapid Diagnostic Testing for Health Care and Environmental Monitoring

The importance of diversity for excellence in science

Machine-in-the-Loop Rewriting for Creative Image Captioning

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES