Search

Although conceptual assessment tests are frequently administered in a pre/post-semester fashion, there are inherent issues with this paradigm. Specifically, education researchers and instructors have limited ability to observe the progression of student conceptual understanding throughout the course. Furthermore, instructors are limited in the usefulness of the feedback they can give to the students involved. To address these issues, we propose the use of computerized adaptive testing (CAT) and increasing the frequency of CAT-based assessments during the course, while reducing the test length per administration, thus keeping or decreasing the total number of test items administered throughout the course. The feasibility of this idea depends on how far the test length per administration can be reduced without compromising the test accuracy and precision. Specifically, the overall test length is desired to be shorter than when the full assessment is administered as a pretest and subsequent post-test. To achieve this goal, we developed a CAT algorithm that we call Chain-CAT. This algorithm sequentially links the results of each CAT administration using collateral information. We developed the Chain-CAT algorithm using the items of the Force Concept Inventory (FCI) and analyzed the efficiency by numerical simulations. We found that collateral information significantly improved the test efficiency, and the overall test length could be shorter than the pre-post method. Without constraints for item balancing and exposure control, simulation results indicated that the efficiency of Chain-CAT is comparable to that of the pre-post method even if the length of each CAT administration is only 5 items and the CAT is administered 9 times throughout the semester. (To continue, see text.)

The theory of a free spin-2 field on Minkowski spacetime has 1-form and $(d-3)$-form symmetries associated with conserved currents formed by contractions of the linearised Riemann tensor with conformal Killing-Yano 2-forms. We show that a subset of these can be interpreted as Noether currents for specific shift symmetries of the graviton that involve a Killing vector and a closed 1-form parameter. We give a systematic method to gauge these 1-form symmetries by coupling the currents to background gauge fields and introducing a particular set of counter-terms involving the background fields. The simultaneous gauging of certain pairs of 1-form and $(d-3)$-form symmetries is obstructed by the presence of mixed 't Hooft anomalies. The anomalous pairs of symmetries are those which are related by gravitational duality. The implications of these anomalies are discussed., Comment: 25 pages + appendices

Mental-health therapy involves a complex conversation flow in which patients and therapists continuously negotiate what should be talked about next. For example, therapists might try to shift the conversation's direction to keep the therapeutic process on track and avoid stagnation, or patients might push the discussion towards issues they want to focus on. How do such patient and therapist redirections relate to the development and quality of their relationship? To answer this question, we introduce a probabilistic measure of the extent to which a certain utterance immediately redirects the flow of the conversation, accounting for both the intention and the actual realization of such a change. We apply this new measure to characterize the development of patient-therapist relationships over multiple sessions in a very large, widely-used online therapy platform. Our analysis reveals that (1) patient control of the conversation's direction generally increases relative to that of the therapist as their relationship progresses; and (2) patients who have less control in the first few sessions are significantly more likely to eventually express dissatisfaction with their therapist and terminate the relationship., Comment: To appear in the Proceedings of EMNLP (Findings) 2024. Code available at https://convokit.cornell.edu

The efficacy of deep learning-based Computer-Aided Diagnosis (CAD) methods for skin diseases relies on analyzing multiple data modalities (i.e., clinical+dermoscopic images, and patient metadata) and addressing the challenges of multi-label classification. Current approaches tend to rely on limited multi-modal techniques and treat the multi-label problem as a multiple multi-class problem, overlooking issues related to imbalanced learning and multi-label correlation. This paper introduces the innovative Skin Lesion Classifier, utilizing a Multi-modal Multi-label TransFormer-based model (SkinM2Former). For multi-modal analysis, we introduce the Tri-Modal Cross-attention Transformer (TMCT) that fuses the three image and metadata modalities at various feature levels of a transformer encoder. For multi-label classification, we introduce a multi-head attention (MHA) module to learn multi-label correlations, complemented by an optimisation that handles multi-label and imbalanced learning problems. SkinM2Former achieves a mean average accuracy of 77.27% and a mean diagnostic accuracy of 77.85% on the public Derm7pt dataset, outperforming state-of-the-art (SOTA) methods., Comment: Accepted by WACV2025

Endometriosis, affecting about 10% of individuals assigned female at birth, is challenging to diagnose and manage. Diagnosis typically involves the identification of various signs of the disease using either laparoscopic surgery or the analysis of T1/T2 MRI images, with the latter being quicker and cheaper but less accurate. A key diagnostic sign of endometriosis is the obliteration of the Pouch of Douglas (POD). However, even experienced clinicians struggle with accurately classifying POD obliteration from MRI images, which complicates the training of reliable AI models. In this paper, we introduce the Human-AI Collaborative Multi-modal Multi-rater Learning (HAICOMM) methodology to address the challenge above. HAICOMM is the first method that explores three important aspects of this problem: 1) multi-rater learning to extract a cleaner label from the multiple "noisy" labels available per training sample; 2) multi-modal learning to leverage the presence of T1/T2 MRI images for training and testing; and 3) human-AI collaboration to build a system that leverages the predictions from clinicians and the AI model to provide more accurate classification than standalone clinicians and AI models. Presenting results on the multi-rater T1/T2 MRI endometriosis dataset that we collected to validate our methodology, the proposed HAICOMM model outperforms an ensemble of clinicians, noisy-label learning models, and multi-rater learning methods.

Linearised gravity has a global symmetry under which the graviton is shifted by a symmetric tensor satisfying a certain flatness condition. There is also a dual symmetry that can be associated with a global shift symmetry of the dual graviton theory. The corresponding conserved charges are shown to satisfy a centrally-extended algebra. We discuss the gauging of these global symmetries, finding an obstruction to the simultaneous gauging of both symmetries which we interpret as a mixed 't Hooft anomaly for the ungauged theory. We discuss the implications of this, analogous to those resulting from a similar structure in Maxwell theory, and interpret the graviton and dual graviton as Nambu-Goldstone modes for these shift symmetries., Comment: 30 pages

The contribution of the magnetic field to the formation of high-mass stars is poorly understood. We report the high-angular resolution ($\sim0.3^{\prime\prime}$, 870 au) map of the magnetic field projected on the plane of the sky (B$_\mathrm{POS}$) towards the high-mass star forming region G333.46$-$0.16 (G333), obtained with the Atacama Large Millimeter/submillimeter Array (ALMA) at 1.2 mm as part of the Magnetic Fields in Massive Star-forming Regions (MagMaR) survey. The B$_\mathrm{POS}$ morphology found in this region is consistent with a canonical ``hourglass'' which suggest a dynamically important field. This region is fragmented into two protostars separated by $\sim1740$ au. Interestingly, by analysing H$^{13}$CO$^{+}$ ($J=3-2$) line emission, we find no velocity gradient over the extend of the continuum which is consistent with a strong field. We model the B$_\mathrm{POS}$, obtaining a marginally supercritical mass-to-flux ratio of 1.43, suggesting an initially strongly magnetized environment. Based on the Davis-Chandrasekhar-Fermi method, the magnetic field strength towards G333 is estimated to be 5.7 mG. The absence of strong rotation and outflows towards the central region of G333 suggests strong magnetic braking, consistent with a highly magnetized environment. Our study shows that despite being a strong regulator, the magnetic energy fails to prevent the process of fragmentation, as revealed by the formation of the two protostars in the central region.

We study the topological dynamics of the action of an acylindrically hyperbolic group on the space of its infinite index convex cocompact subgroups by conjugation. We show that, for any suitable probability measure $\mu$, random walks with respect to $\mu$ will produce elements with strong mixing properties for this action asymptotically almost surely. In particular, when the group has no finite normal subgroups this implies that the action is highly topologically transitive. Along the way, we prove technical results about convex cocompact subgroups which allow us to extend some results on random walks of Abbott and the first author.

Developing theoretical understanding of complex reactions and processes at interfaces requires using methods that go beyond semilocal density functional theory to accurately describe the interactions between solvent, reactants and substrates. Methods based on many-body perturbation theory, such as the random phase approximation (RPA), have previously been limited due to their computational complexity. However, this is now a surmountable barrier due to the advances in computational power available, in particular through modern GPU-based supercomputers. In this work, we describe the implementation of RPA calculations within BerkeleyGW and show its favorable computational performance on large complex systems relevant for catalysis and electrochemistry applications. Our implementation builds off of the static subspace approximation which, by employing a compressed representation of the frequency dependent polarizability, enables the evaluation of the RPA correlation energy with significant acceleration and systematically controllable accuracy. We find that the computational cost of calculating the RPA correlation energy scales only linearly with system size for systems containing up to 50 thousand bands, and is expected to scale quadratically thereafter. We also show excellent strong scaling results across several supercomputers, demonstrating the performance and portability of this implementation., Comment: 10 pages, 5 figures

Safety alignment is the key to guiding the behaviors of large language models (LLMs) that are in line with human preferences and restrict harmful behaviors at inference time, but recent studies show that it can be easily compromised by finetuning with only a few adversarially designed training examples. We aim to measure the risks in finetuning LLMs through navigating the LLM safety landscape. We discover a new phenomenon observed universally in the model parameter space of popular open-source LLMs, termed as "safety basin": randomly perturbing model weights maintains the safety level of the original aligned model in its local neighborhood. Our discovery inspires us to propose the new VISAGE safety metric that measures the safety in LLM finetuning by probing its safety landscape. Visualizing the safety landscape of the aligned model enables us to understand how finetuning compromises safety by dragging the model away from the safety basin. LLM safety landscape also highlights the system prompt's critical role in protecting a model, and that such protection transfers to its perturbed variants within the safety basin. These observations from our safety landscape research provide new insights for future work on LLM safety community.

Linearised gravity has magnetic charges carried by (linearised) Kaluza-Klein monopoles. A gauge-invariant expression is found for these charges that is similar to Penrose's gauge-invariant expression for the ADM charges. A systematic search is made for other gauge-invariant charges., Comment: 47 pages. v2: minor revisions. v3: fixed typographical error

In multi-modal learning, some modalities are more influential than others, and their absence can have a significant impact on classification/segmentation accuracy. Hence, an important research question is if it is possible for trained multi-modal models to have high accuracy even when influential modalities are absent from the input data. In this paper, we propose a novel approach called Meta-learned Cross-modal Knowledge Distillation (MCKD) to address this research question. MCKD adaptively estimates the importance weight of each modality through a meta-learning process. These dynamically learned modality importance weights are used in a pairwise cross-modal knowledge distillation process to transfer the knowledge from the modalities with higher importance weight to the modalities with lower importance weight. This cross-modal knowledge distillation produces a highly accurate model even with the absence of influential modalities. Differently from previous methods in the field, our approach is designed to work in multiple tasks (e.g., segmentation and classification) with minimal adaptation. Experimental results on the Brain tumor Segmentation Dataset 2018 (BraTS2018) and the Audiovision-MNIST classification dataset demonstrate the superiority of MCKD over current state-of-the-art models. Particularly in BraTS2018, we achieve substantial improvements of 3.51\% for enhancing tumor, 2.19\% for tumor core, and 1.14\% for the whole tumor in terms of average segmentation Dice score.

We take a fresh look at the relation between generalised K\"ahler geometry and $N=(2,2)$ supersymmetric sigma models in two dimensions formulated in terms of $(2,2)$ superfields. Dual formulations in terms of different kinds of superfield are combined to give a formulation with a doubled target space and both the original superfield and the dual superfield. For K\"ahler geometry, we show that this doubled geometry is Donaldson's deformation of the holomorphic cotangent bundle of the original K\"ahler manifold. This doubled formulation gives an elegant geometric reformulation of the equations of motion. We interpret the equations of motion as the intersection of two Lagrangian submanifolds (or of a Lagrangian submanifold with an isotropic one) in the infinite dimensional symplectic supermanifold which is the analogue of phase space. We then consider further extensions of this formalism, including one in which the geometry is quadrupled, and discuss their geometry., Comment: 44 pages, final published version

Non-periodic folding of periodic crease patterns paves the way to novel nonlinear phenomena that cannot be feasible through periodic folding. This paper focuses on the non-periodic folding of recursive crease patterns generalized from Spidron. Although it is known that the Spidron has a $1$-DOF isotropic rigid folding motion, its general kinematics and dependence on the crease pattern remain unclear. Using the kinematics of a single unit cell of the Spidron and the recursive construction of the folded state of multiple unit cells, we consider the folding of the Spidron that is not necessarily isotropic. We found that as the number of unit cells increases, the non-periodic folding is restricted and the isotropic folding becomes dominant. Then, we analyze the three kinds of isotropic folding modes by constructing $1$-dimensional dynamical systems governing each of them. We show that these systems can possess different recursive natures depending on folding modes even in an identical crease pattern. Furthermore, we show their novel nonlinear nature, including the period-doubling cascade leading to the emergence of chaos., Comment: 16 pages, 8 figures. To Appear in Origami8: Proceedings of the 8th International Meeting on Origami in Science, Mathematics and Education (8OSME), Melbourne, August 16-18, 2024

We use freely acting asymmetric orbifolds of type IIB string theory to construct a class of theories in five dimensions with eight supercharges whose moduli spaces for vector multiplets and hypermultiplets can be determined exactly. We argue that no quantum corrections to these moduli spaces arise. We focus on examples in which all moduli are in the NS-NS sector, while all fields from the R-R sector become massive. The full symmetry group of the moduli space is then determined by the subgroup of the T-duality group that survives the orbifold action. We illustrate this for freely acting orbifolds of type IIB string theory on $T^5$ with $0,1$ or $2$ hypermultiplets., Comment: 38 pages

We present 870 um polarimetric observations toward 61 protostars in the Orion molecular clouds, with ~400 au (1") resolution using the Atacama Large Millimeter/submillimeter Array. We successfully detect dust polarization and outflow emission in 56 protostars, in 16 of them the polarization is likely produced by self-scattering. Self-scattering signatures are seen in several Class 0 sources, suggesting that grain growth appears to be significant in disks at earlier protostellar phases. For the rest of the protostars, the dust polarization traces the magnetic field, whose morphology can be approximately classified into three categories: standard-hourglass, rotated-hourglass (with its axis perpendicular to outflow), and spiral-like morphology. 40.0% (+-3.0%) of the protostars exhibit a mean magnetic field direction approximately perpendicular to the outflow on several 100--1000 au scales. However, in the remaining sample, this relative orientation appears to be random, probably due to the complex set of morphologies observed. Furthermore, we classify the protostars into three types based on the C17O (3--2) velocity envelope's gradient: perpendicular to outflow, non-perpendicular to outflow, and unresolved gradient (<1.0~km/s/arcsec). In protostars with a velocity gradient perpendicular to outflow, the magnetic field lines are preferentially perpendicular to outflow, most of them exhibit a rotated hourglass morphology, suggesting that the magnetic field has been overwhelmed by gravity and angular momentum. Spiral-like magnetic fields are associated with envelopes having large velocity gradients, indicating that the rotation motions are strong enough to twist the field lines. All of the protostars with a standard-hourglass field morphology show no significant velocity gradient due to the strong magnetic braking., Comment: 12 pages, 5 figures (16 pages and 7 figures for the supplementary materials) Accepted for publication on ApJ Letters

Covariant conserved 2-form currents for linearised gravity are constructed by contracting the linearised curvature with conformal Killing-Yano tensors. The corresponding conserved charges were originally introduced by Penrose and have recently been interpreted as the generators of generalised symmetries of the graviton. We introduce an off-shell refinement of these charges and find the relation between these improved Penrose charges and the linearised version of the ADM momentum and angular momentum. If the graviton field is globally well-defined on a background Minkowski space then some of the Penrose charges give the momentum and angular momentum while the remainder vanish. We consider the generalisation in which the graviton has Dirac string singularities or is defined locally in patches, in which case the conventional ADM expressions are not invariant under the graviton gauge symmetry in general. We modify them to render them gauge-invariant and show that the Penrose charges give these modified charges plus certain magnetic gravitational charges. We discuss properties of the Penrose charges, generalise to toroidal Kaluza-Klein compactifications and check our results in a number of examples., Comment: 37 pages + appendices

The recent discovery of ``ultra-hot'' ($P < 1$ day) Neptunes has come as a surprise: some of these planets have managed to retain gaseous envelopes despite being close enough to their host stars to trigger strong photoevaporation and/or Roche lobe overflow. Here, we investigate atmospheric escape in LTT 9779b, an ultra-hot Neptune with a volatile-rich envelope. We observed two transits of this planet using the newly-commissioned WINERED spectrograph ($R\sim68,000$) on the 6.5 m Clay/Magellan II Telescope, aiming to detect an extended upper atmosphere in the He 10830 A triplet. We found no detectable planetary absorption: in a 0.75 A passband centered on the triplet, we set a 2$\sigma$ upper limit of 0.12% ($\delta R_p/H < 14$) and a 3$\sigma$ upper limit of 0.20% ($\delta R_p/H < 22$). Using a H/He isothermal Parker wind model, we found corresponding 95% and 99.7% upper limits on the planetary mass-loss rate of $\dot{M} < 10^{10.03}$ g s$^{-1}$ and $\dot{M} < 10^{11.11}$ g s$^{-1}$ respectively, smaller than predicted by outflow models even considering the weak stellar XUV emission. The low evaporation rate is plausibly explained by a metal-rich envelope, which would decrease the atmospheric scale height and increase the cooling rate of the outflow. This hypothesis is imminently testable: if metals commonly weaken planetary outflows, then we expect that \textit{JWST} will find high atmospheric metallicities for small planets that have evaded detection in He 10830 A., Comment: 11 pages, 3 figures, accepted for publication in ApJL

A novel type of calorimeter based on grains of inorganic scintillating crystal readout by wave length shifting fibers is proposed. The concept and main features as well as the prototype design are introduced and the first results obtained using cosmic rays are presented. The number of photo-electrons generated by cosmic rays muons in the prototype detector is estimated to be of the order of 10000 photo-electrons per GeV, validating the concept of this next-generation shashlik calorimeter., Comment: 11 pages, 7 figures

Addressing public safety effectively requires incorporating diverse stakeholder perspectives, particularly those of the community, which are often underrepresented compared to other stakeholders. This study presents a comprehensive analysis of the community's general public safety concerns, their view of existing surveillance technologies, and their perception of AI-driven solutions for enhancing safety in urban environments, focusing on Charlotte, NC. Through a survey approach, including in-person surveys conducted in August and September 2023 with 410 participants, this research investigates demographic factors such as age, gender, ethnicity, and educational level to gain insights into public perception and concerns toward public safety and possible solutions. Based on the type of dependent variables, we utilized different statistical and significance analyses, such as logit regression and ordinal logistic regression, to explore the effects of demographic factors on the various dependent variables. Our results reveal demographic differences in public safety concerns. Younger females tend to feel less secure yet trust existing video surveillance systems, whereas older, educated individuals are more concerned about violent crimes in malls. Additionally, attitudes towards AI-driven surveillance differ: older Black individuals demonstrate support for it despite having concerns about data privacy, while educated females show a tendency towards skepticism., Comment: 12 pages, 4 figures

4-Dimensional Einstein-Gauss-Bonnet (4DEGB) gravity has garnered significant attention in the last few years as a phenomenological competitor to general relativity. We consider the theoretical and observational implications of this theory in both the early and late universe, (re-)deriving background and perturbation equations and constraining its characteristic parameters with data from cosmological probes. Our investigation surpasses the scope of previous studies by incorporating non-flat spatial sections. We explore consequences of 4DEGB on the sound and particle horizons in the very early universe, and demonstrate that 4DEGB can provide an independent solution to the horizon problem for some values of its characteristic parameter $\alpha$. Finally, we constrain an unexplored regime of this theory in the limit of small coupling $\alpha$ (empirically supported in the post-Big Bang Nucleosynthesis era by prior constraints). This version of 4DEGB includes a geometric term that resembles dark radiation at the background level, but whose influence on the perturbed equations is qualitatively distinct from that of standard forms of dark radiation. In this limit, only one beyond-$\Lambda$CDM degree of freedom persists, which we denote as $\tilde{\alpha}_C$. Our analysis yields the estimate $\tilde{\alpha}_C = (-9 \pm 6) \times 10^{-6}$ thereby providing a new constraint of a previously untested sector of 4DEGB., Comment: 17 pages, 5 figures (excluding appendices). Updated to reflect JCAP author-accepted version

Symmetries and conserved charges are investigated for linearised gravity and its dual formulation in terms of the dual graviton field. Conserved charges are constructed for the dual graviton theory as Noether charges associated with invariances of the dual graviton. These invariances arise for gauge parameters that are certain generalised Killing tensors. These electric-type charges for the dual graviton are then shown to give magnetic-type charges for the graviton. One of the magnetic-type gravitational charges obtained in this way is the linearisation of the gravitational charge in $d\ge 5$ dimensions that arises as a central charge in the supersymmetry algebra and is carried by Kaluza-Klein monopoles. Solutions of linearised gravity carrying the magnetic gravitational charges are discussed. The application of the approach used here to other gauge theories is discussed., Comment: 51 Pages. Improved discussion of Noether charges and of the topological nature of the magnetic charges

The amount of vapor in the impact-generated protolunar disk carries implications for the dynamics, devolatilization, and moderately volatile element (MVE) isotope fractionation during lunar formation. The equation of state (EoS) used in simulations of the giant impact is required to calculate the vapor mass fraction (VMF) of the modeled protolunar disk. Recently, a new version of M-ANEOS ("Stewart M-ANEOS") was released with an improved treatment of heat capacity and expanded experimental Hugoniot. Here, we compare this new M-ANEOS version with a previous version ("N-SPH M-ANEOS") and assess the resulting differences in smoothed particle hydrodynamics (SPH) simulations. We find that Stewart M-ANEOS results in cooler disks with smaller values of VMF and results in differences in disk mass that are dependent on the initial impact angle. We also assess the implications of the minimum "cutoff" density ($\rho_{c}$), similar to a maximum smoothing length, that is set as a fast-computing alternative to an iteratively calculated smoothing length. We find that the low particle resolution of the disk typically results in $>40\%$ of disk particles falling to $\rho_c$, influencing the dynamical evolution and VMF of the disk. Our results show that choice of EoS, $\rho_{c}$, and particle resolution can cause the VMF and disk mass to vary by tens of percent. Moreover, small values of $\rho_{c}$ produce disks that are prone to numerical instability and artificial shocks. We recommend that future giant impact SPH studies review smoothing methods and ensure the thermodynamic stability of the disk over simulated time., Comment: To be published in The Planetary Science Journal, 51 pages, 34 figures, 4 tables

Manually grading D3 data visualizations is a challenging endeavor, and is especially difficult for large classes with hundreds of students. Grading an interactive visualization requires a combination of interactive, quantitative, and qualitative evaluation that are conventionally done manually and are difficult to scale up as the visualization complexity, data size, and number of students increase. We present VisGrader, a first-of-its kind automatic grading method for D3 visualizations that scalably and precisely evaluates the data bindings, visual encodings, interactions, and design specifications used in a visualization. Our method enhances students learning experience, enabling them to submit their code frequently and receive rapid feedback to better inform iteration and improvement to their code and visualization design. We have successfully deployed our method and auto-graded D3 submissions from more than 4000 students in a visualization course at Georgia Tech, and received positive feedback for expanding its adoption.

Deep Learning models, such as those used in an autonomous vehicle are vulnerable to adversarial attacks where an attacker could place an adversarial object in the environment, leading to mis-classification. Generating these adversarial objects in the digital space has been extensively studied, however successfully transferring these attacks from the digital realm to the physical realm has proven challenging when controlling for real-world environmental factors. In response to these limitations, we introduce REVAMP, an easy-to-use Python library that is the first-of-its-kind tool for creating attack scenarios with arbitrary objects and simulating realistic environmental factors, lighting, reflection, and refraction. REVAMP enables researchers and practitioners to swiftly explore various scenarios within the digital realm by offering a wide range of configurable options for designing experiments and using differentiable rendering to reproduce physically plausible adversarial objects. We will demonstrate and invite the audience to try REVAMP to produce an adversarial texture on a chosen object while having control over various scene parameters. The audience will choose a scene, an object to attack, the desired attack class, and the number of camera positions to use. Then, in real time, we show how this altered texture causes the chosen object to be mis-classified, showcasing the potential of REVAMP in real-world scenarios. REVAMP is open-source and available at https://github.com/poloclub/revamp.

Prior work has shown that analyzing the use of first-person singular pronouns can provide insight into individuals' mental status, especially depression symptom severity. These findings were generated by counting frequencies of first-person singular pronouns in text data. However, counting doesn't capture how these pronouns are used. Recent advances in neural language modeling have leveraged methods generating contextual embeddings. In this study, we sought to utilize the embeddings of first-person pronouns obtained from contextualized language representation models to capture ways these pronouns are used, to analyze mental status. De-identified text messages sent during online psychotherapy with weekly assessment of depression severity were used for evaluation. Results indicate the advantage of contextualized first-person pronoun embeddings over standard classification token embeddings and frequency-based pronoun analysis results in predicting depression symptom severity. This suggests contextual representations of first-person pronouns can enhance the predictive utility of language used by people with depression symptoms., Comment: Accepted: AMIA Annual Symposium 2023. To appear as: Ren X, Burkhardt H, Are\'an P, Hull T, Cohen T. Deep Representations of First-person Pronouns for Prediction of Depression Symptom Severity. AMIA Annual Symposium Proceedings 2023. American Medical Informatics Association

The problem of missing modalities is both critical and non-trivial to be handled in multi-modal models. It is common for multi-modal tasks that certain modalities contribute more compared to other modalities, and if those important modalities are missing, the model performance drops significantly. Such fact remains unexplored by current multi-modal approaches that recover the representation from missing modalities by feature reconstruction or blind feature aggregation from other modalities, instead of extracting useful information from the best performing modalities. In this paper, we propose a Learnable Cross-modal Knowledge Distillation (LCKD) model to adaptively identify important modalities and distil knowledge from them to help other modalities from the cross-modal perspective for solving the missing modality issue. Our approach introduces a teacher election procedure to select the most ``qualified'' teachers based on their single modality performance on certain tasks. Then, cross-modal knowledge distillation is performed between teacher and student modalities for each task to push the model parameters to a point that is beneficial for all tasks. Hence, even if the teacher modalities for certain tasks are missing during testing, the available student modalities can accomplish the task well enough based on the learned knowledge from their automatically elected teacher modalities. Experiments on the Brain Tumour Segmentation Dataset 2018 (BraTS2018) shows that LCKD outperforms other methods by a considerable margin, improving the state-of-the-art performance by 3.61% for enhancing tumour, 5.99% for tumour core, and 3.76% for whole tumour in terms of segmentation Dice score.

In the U.S. inpatient payment system, the Diagnosis-Related Group (DRG) is pivotal, but its assignment process is inefficient. The study introduces DRG-LLaMA, an advanced large language model (LLM) fine-tuned on clinical notes to enhance DRGs assignment. Utilizing LLaMA as the foundational model and optimizing it through Low-Rank Adaptation (LoRA) on 236,192 MIMIC-IV discharge summaries, our DRG-LLaMA-7B model exhibited a noteworthy macro-averaged F1 score of 0.327, a top-1 prediction accuracy of 52.0%, and a macro-averaged Area Under the Curve (AUC) of 0.986, with a maximum input token length of 512. This model surpassed the performance of prior leading models in DRG prediction, showing a relative improvement of 40.3% and 35.7% in macro-averaged F1 score compared to ClinicalBERT and CAML, respectively. Applied to base DRG and complication or comorbidity (CC)/major complication or comorbidity (MCC) prediction, DRG-LLaMA achieved a top-1 prediction accuracy of 67.8% and 67.5%, respectively. Additionally, our findings indicate that DRG-LLaMA's performance correlates with increased model parameters and input context lengths.

Flat origami refers to the folding of flat, zero-curvature paper such that the finished object lies in a plane. Mathematically, flat origami consists of a continuous, piecewise isometric map $f:P\subseteq\mathbb{R}^2\to\mathbb{R}^2$ along with a layer ordering $\lambda_f:P\times P\to \{-1,1\}$ that tracks which points of $P$ are above/below others when folded. The set of crease lines that a flat origami makes (i.e., the set on which the mapping $f$ is non-differentiable) is called its \textit{crease pattern}. Flat origami mappings and their layer orderings can possess surprisingly intricate structure. For instance, determining whether or not a given straight-line planar graph drawn on $P$ is the crease pattern for some flat origami has been shown to be an NP-complete problem, and this result from 1996 led to numerous explorations in computational aspects of flat origami. In this paper we prove that flat origami, when viewed as a computational device, is Turing complete. We do this by showing that flat origami crease patterns with \textit{optional creases} (creases that might be folded or remain unfolded depending on constraints imposed by other creases or inputs) can be constructed to simulate Rule 110, a one-dimensional cellular automaton that was proven to be Turing complete by Matthew Cook in 2004.

Deep neural networks (DNNs) are increasingly proposed as models of human vision, bolstered by their impressive performance on image classification and object recognition tasks. Yet, the extent to which DNNs capture fundamental aspects of human vision such as color perception remains unclear. Here, we develop novel experiments for evaluating the perceptual coherence of color embeddings in DNNs, and we assess how well these algorithms predict human color similarity judgments collected via an online survey. We find that state-of-the-art DNN architectures $-$ including convolutional neural networks and vision transformers $-$ provide color similarity judgments that strikingly diverge from human color judgments of (i) images with controlled color properties, (ii) images generated from online searches, and (iii) real-world images from the canonical CIFAR-10 dataset. We compare DNN performance against an interpretable and cognitively plausible model of color perception based on wavelet decomposition, inspired by foundational theories in computational neuroscience. While one deep learning model $-$ a convolutional DNN trained on a style transfer task $-$ captures some aspects of human color perception, our wavelet algorithm provides more coherent color embeddings that better predict human color judgments compared to all DNNs we examine. These results hold when altering the high-level visual task used to train similar DNN architectures (e.g., image classification versus image segmentation), as well as when examining the color embeddings of different layers in a given DNN architecture. These findings break new ground in the effort to analyze the perceptual representations of machine learning algorithms and to improve their ability to serve as cognitively plausible models of human vision. Implications for machine learning, human perception, and embodied cognition are discussed., Comment: 22 pages, 8 figures + SI Appendix; to appear in Cognition