liangzid/robench2024b_all_setq-bioSCP-s-50

shuffled_text stringlengths 276 2.25k	A stringclasses 6 values	B stringclasses 6 values	C stringclasses 6 values	D stringclasses 6 values	label stringclasses 4 values
A: In total in this dataset there are 220,177 images of 50 x 50 pixels in three colorsB: The ratio between the two classes is 70 % normal and 30 % tumor for this dataset (Figure 2). C: Images classified as tumor by a pathologist are labelled as 1 while normal tissue images are labelled 0	ACB	CAB	BCA	CBA	Selection 1
A: We could however allow for more general initial conditions, as long as they lead to an asymptotic ESC within finitely many steps of the ln⁡K𝐾\ln Kroman_ln italic_K-algorithm. B: Both results assume that the system starts out in an asymptotic ESCC: These are the natural initial conditions, particularly when a first transition between asymptotic ESCs has already occurred	CBA	BCA	BAC	CAB	Selection 4
A: To date, a multitude of mathematical models describing infectious diseases through differential equations have been formulated and scrutinized to understand the dynamics of infection spread, exemplified by research on [1, 2, 3, 4]B: Recently, the mathematical modeling of the COVID-19 pandemic has captivated the attention of numerous experts, including mathematicians, scientists, epidemiologists, pharmacists, and chemistsC: The outcomes of these endeavors have yielded several noteworthy and crucial results, as highlighted in the works of [4, 5, 6, 7, 8]	CBA	ABC	ACB	BAC	Selection 2
A: (2010) reported a heritability of 0.42 for default-mode network (DMN) in an extendedB: Glahn et alC: One of the earliest papers on functional brain activation in twins is based on the resting-state EEG (Lykken et al., 1982), where they observed high twin correlation in MZ-twins on EEG spectra	BCA	ABC	CBA	BCA	Selection 3
A: Therefore, in this note we do not differentiate between the two cell phenotypes and focus instead on therapy effects. We consider tumor cells interacting with their environment, whose main physical components are the anisotropic brain tissue as well as brain-own and tumor-generated vasculature. We also modeled explicitly the evolution of VEGF controlling angiogenesis and the necrotic component of the neoplasm, which is relevant for assessing the tumor stage and for the segmentation needed for treatment planningB: The model investigates the impact of including microscopic dynamics also in the equation for endothelial cells, which has not been done before, comparing our approach with other two possible descriptions of EC dynamics. Then, we analyze via simulations various therapy approaches and their effects on tumor development and on normal tissue, thereby following treatment schedules addressed in clinical studies available in the literature. Although this has been done in [26] in a rather rudimentary way, we do not include tumor resection as part of the therapy; the problems have already been stated in that reference: it is hardly possible to provide a proper mathematical characterization of the tissue and neoplasm dynamics in the resected region in a continuous manner and the solution infers sharp discontinuities, which are difficult to handle. Considering the real data employed in Subsection 4.3, the simulations showed qualitatively reasonable results, however, a quantitative assessment could not be performed, due to missing data, as that patient’s therapy was stopped. C: The model proposed in this note is a therapy-oriented development of that considered in [7] and aligns to the approaches in [6, 8, 9, 13, 14, 15, 26]. Unlike [7, 15] we do not differentiate here between moving and proliferating cancer cells, but account for a single population of cells forming the tumor. In fact, the analysis in [7] showed (at least in our framework) that the main effect of splitting the tumor population into moving and resting (and proliferating) cells was the slower evolution of tumor mass, but qualitatively it behaved similarly to the situation with one population of moving and proliferating cancer cells	ABC	CBA	BCA	CAB	Selection 3
A: The reduction in the number of sulci in the brains of manatees suggests evolutionary trade-offs in brain gyrificationB: For species with low risks of head impact—such as manatees—a lissencephalic structure, devoid of sulci, might facilitate more direct and efficient neural pathways. Conversely, species like humans, who must balance the risks of head impact with the need for neural efficiency, may benefit from moderate gyrification. C: One possible trade-off could be that sulci may compromise efficient neural connectivity, as illustrated in Fig. 11	ACB	CBA	BCA	CBA	Selection 1
A: Subsequently, synergistic and antagonistic classes (i.e. labels) were defined for each synergy score using thresholding based on the SynergyFinder software’s documentation333https://synergyfinder.fimm.fi/synergy/synfin_docs/#datanal, applying recommended thresholds (≥10absent10\geq 10≥ 10 for synergy and ≤−10absent10\leq-10≤ - 10 for antagonism) individually to the four synergy scores. B: The initial dataset underwent the following steps: Triplets with corresponding identifiers in the gene expression features table were selectedC: Samples with missing values in drug or cell line identifiers, absent synergy scores, or duplicated triplets were filtered out	BCA	CBA	CAB	BAC	Selection 3
A: 2B).B: 2F), there is an optimal number of chromosomes for evolving novel traits (FigC: We found that the skewness of the distribution of genotypes coded in each chromosome is essential to determining the probability of evolving new traits. Because skewness peaked with a finite number of chromosomes (Fig	BAC	CBA	BAC	BCA	Selection 2
A: The curl components are zero for edges that are not a 2-simplex boundary and the entries sum to zero around each node. The harmonic component sums to zero around each node, and it also sums to zero along each 2-simplex. We tested the topological equivalence of female brain networks and male brain networks using the Wasserstein distance (5).B: We further determined if we can detect topological differences in the decomposed components (Figure 2)C: The gradient component sum to zero along any cycles	CAB	BCA	CBA	ACB	Selection 1
A: in the case of populations suffering recurrent bottlenecks [CasanovaMiroJegousse2020] or overlapping generations (Casanova et alB: Others are modifications, such as models with varying population sizes [Mohle2002, KajKrone2003, Freund2020] or diploid reproduction [MohleSagitov2003, BirknerLiuSturm2018], using heuristics close to those of [MohleSagitov2001]. Finally some works use different techniques such as duality, e.gC: preprint arXiv:2210.05819).	BCA	ACB	BAC	ACB	Selection 3
A: To assess the effectiveness of the proposed model in capturing the morphological features of prostate cancer, a random subset of the WSIs was selected from the test set. Heatmaps, as shown in FigB: RGB color maps were used to enhance the clarity, with red representing high attention and blue indicating low attention. These heat maps were overlaid onto the original WSIs. C: 4, were generated using a patch size of 224×224224224224\times 224224 × 224 and 90% overlay. The attention score of each patch was normalized to the range [0, 1], with high scores indicating regions crucial for diagnosis	BCA	BCA	ACB	BCA	Selection 3
A: The dataset was generated by filtering examples from the Protein Data Bank (PDB) corresponding to the same protein but bound to different biomolecules, with additional quality control criteria. For the scope of this work, we only focus on protein pairs where the provided Root Mean Square Deviation (RMSD) of the Cα𝛼\alphaitalic_α carbon atoms between unbound and bound 3D structures is >3.0absent3.0>3.0> 3.0Å, which amounts to 2370 examples in the D3PM dataset. B: Here we utilize the recently proposed D3PM dataset (Peng et al., 2022) that provides protein structures before (apo) and after (holo) binding, covering various types of protein motionsC: The task of modeling conformational changes starting from a given protein structure is largely unexplored, mainly due to the lack of high-quality large datasets	CAB	ABC	BCA	CBA	Selection 4
A: A characteristic equation was also deduced for the special case when neither the growth rate g𝑔gitalic_g nor the mortality rate μ𝜇\muitalic_μ depend on the interaction variable E𝐸Eitalic_E (β𝛽\betaitalic_β on the other hand does). Note however that the linearisation and stability results established in [15] were completely formal, as the Principle of Linearised Stability has not been established for the PDE formulation (1). This is the main reason why in the current work we employ a different formulation of the model. B: On the other hand, in [15] the authors derived a formal linearisation of the model and studied regularity properties of the governing linear semigroupC: We note that the focus in [1, 2] was on numerical approximation of solutions of the hierarchical model	ACB	CBA	BCA	BCA	Selection 2
A: The classical representation of coloured graphs is through their colour classesB: In this section we introduce a different representation of pdCGs that will allow us to deal more efficiently with these objectsC: A detailed example that may help to follow the steps required for the construction of the novel representation is given in Section A of the Appendix.	CAB	BCA	ABC	CAB	Selection 3
A: We standardize cardiac MRI intensity levels using Z-score normalization [7] to eliminate inter-subject variationsB: Furthermore, we detect automatic landmarks which is explained in the next paragraph. We perform affine registration to align the heart regions of different subjects to a target image space. We then carry out in-plane scaling of scans by max-pooling at 2222, 4444, 8888, and 16161616 times and obtain down-sampled resolutions of 128×128128128128\times 128128 × 128, 64×64646464\times 6464 × 64, 32×32323232\times 3232 × 32, and 16×16161616\times 1616 × 16, respectively.C: Cardiac MRI Preprocessing: The preprocessing of cardiac MRI contains (1111) normalization of scans, (2222) automatic landmark detection, (3333) inter-subject registration, and (4444) in-plane downsampling	CAB	ACB	BCA	CBA	Selection 3
A: This paper addresses two challenges posed by the multimodal fusion of transcriptomics and histology: (1) we address the challenge of transcriptomics tokenization by defining biological pathway tokens that encode semantically meaningful and interpretable functions, and (2) we overcome the computational challenge of integrating long multimodal sequences by designing a multimodal Transformer with sparse modality-specific attention patterns. Our model achieves state-of-the-art survival performance when tested on five datasets from TCGAB: In addition, our interpretability framework reveals known and candidate prognostic features. While our interpretability framework enables identifying prognostic features, these findings remain qualitativeC: Future work could focus on interpretability metrics that generalize findings at dataset-level, e.g., with quantitative morphological characterizations of specific pathways. In addition, our findings suggest that including patch-to-patch interactions does not lead to improved performance. Nonetheless, the absence of a performance boost should not be an evidence that patch-to-patch interactions are unnecessary, but rather that modeling such interactions is a challenging problem that remains to be solved.	BCA	CAB	ABC	BCA	Selection 3
A: where we have applied Stirling’s approximation [30] to derive an asymptotic expression for the large compartment number, ν≫1much-greater-than𝜈1\nu\gg 1italic_ν ≫ 1, limitB: Even for ν=1𝜈1\nu=1italic_ν = 1, the asymptotic expression produces excellent agreement with simulation results.C: In fig. 2a, we compare both the exact and asymptotic expressions for the stationary variance to a numerical approximation produced through repeated simulation of the SDE	ABC	ACB	ABC	BCA	Selection 2
A: For each dataset, it establishes the preferred metric for evaluation to enable consistent comparison across models. We describe each dataset selected to evaluate our model.B: MoleculeNet is a benchmark set used to evaluate machine learning techniques [9]C: It curate’s quantum mechanics, physical chemistry, biophysics, and physiology datasets	CAB	ACB	ABC	ACB	Selection 1
A: In the quenched regime, the network changes very slowly compared to the evolution of the epidemicB: Indeed, as we will show in Section I.3, the quenched regime assumes that, before each topology update, the epidemic has almost reached its equilibrium. C: Quenched processes are well approximated with processes on static networks and therefore well studied over the last two decades	ABC	ABC	BAC	ACB	Selection 4
A: We allow degenerate vertices with in-degree 2222 as well as out-degree 2B: The cover 𝒞𝒞\mathcal{C}caligraphic_C for N𝑁Nitalic_N then consists of sets of size 1 or 2, and each integer appearing in 𝒞𝒞\mathcal{C}caligraphic_C appears either once, if it is a tree vertex (in-degree 1), or twice if it is a reticulation (in-degree 2). C: Suppose N𝑁Nitalic_N is a binary tree-based network	ABC	BCA	ABC	BAC	Selection 2
A: As pointed out by Barton et al. (2004), the Moran model with finite N𝑁Nitalic_N is reversible, meaning that at stationarity the time-reversed process is the same (in distribution) as the forward-time Moran processB: This is not a property of the Wright-Fisher model with finite N𝑁Nitalic_N but does hold for their shared diffusion limit (1) with stationary density (2); see for instance Millet et al. (1989) for why this holds. Figure 1 gives an illustration of a genealogy with mutations and allele frequencies varying backward in time.C: Even if the sample frequencies are known, the allele frequencies in the population are unknown. A key feature of this method is to model allele-frequency trajectories backward in time	CBA	BAC	BCA	ACB	Selection 3
A: We tackle this problem by assuming that each read is coming from the one (e.gB: ’maternal’) chromosome with probability w𝑤witalic_w and from the other chromosome (e.gC: paternal) with probability 1−w1𝑤1-w1 - italic_w, where the balance between w𝑤witalic_w and 1−w1𝑤1-w1 - italic_w reflects BAD. This is done naturally with the mixture distribution:	ABC	BAC	BAC	BCA	Selection 1
A: We set Ci,k+1=Isubscript𝐶𝑖𝑘1𝐼C_{i,k+1}=Iitalic_C start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_I and τi,k+1=τi,k+Zi,ksubscript𝜏𝑖𝑘1subscript𝜏𝑖𝑘subscript𝑍𝑖𝑘\tau_{i,k+1}=\tau_{i,k}+Z_{i,k}italic_τ start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_τ start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPT + italic_Z start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPTB: vaccinatedC: Otherwise, the individual is vaccinated before being reinfected, and we set Ci,k+1=Ssubscript𝐶𝑖𝑘1𝑆C_{i,k+1}=Sitalic_C start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_S, and τi,k+1=τi,k+TV,i,ksubscript𝜏𝑖𝑘1subscript𝜏𝑖𝑘subscript𝑇𝑉𝑖𝑘\tau_{i,k+1}=\tau_{i,k}+T_{V,i,k}italic_τ start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_τ start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPT + italic_T start_POSTSUBSCRIPT italic_V , italic_i , italic_k end_POSTSUBSCRIPT.	CAB	BCA	CBA	BAC	Selection 4
A: acknowledges Iniziativa PNC0000002-DARE - Digital Lifelong Prevention. S.A., F.F., and A.M. also acknowledge the support by the Italian Ministry of University and Research (project funded by the European Union - Next Generation EU: “PNRR Missione 4 Componente 2, “Dalla ricerca all’impresa”, Investimento 1.4, Progetto CN00000033”).B: Acknowledgements. We wish to acknowledge Jacopo Grilli and Davide Bernardi for critical reading of the manuscript and useful discussions. F.FC: thanks Matteo Guardiani and the Information Field Theory group at the Max-Planck Institute for Astrophysics for their hospitality and helpful comments. S.S	CBA	CAB	ACB	BCA	Selection 2
A: In the non-stationary case, the results are possible, but more involvingB: The above results pertain to the stationary caseC: The non-stationary queue length distribution can be computed using renewal theory provided one can compute the distribution of the forward recurrence time (the time until the next arrival) [72], whereas the moments can be computed recursively without the knowledge of this distribution [71].	BAC	CBA	CBA	CAB	Selection 1
A: The remainder of this paper consists of two sectionsB: The next section contains the proofs of our theoretical results from Section 2C: In Section 4 we provide a numerical example of an MTBDP where the rates depend on the number of particles of each type that are present in the system.	BAC	ABC	ACB	ACB	Selection 2
A: As a result, the RGCN encoder might struggle to capture complex patterns and may not fully leverage the underlying protein data, potentially leading to suboptimal performance.B: One limitation of our proposed Prot2Text model is that the RGCN encoder is not pretrainedC: Unlike the ESM encoder, which benefits from pretraining on a large corpus, the RGCN encoder lacks this initial knowledge	ABC	CAB	CBA	BAC	Selection 2
A: measured spectral radius rsimsubscript𝑟simr_{\mathrm{sim}}italic_r start_POSTSUBSCRIPT roman_sim end_POSTSUBSCRIPT. Same excitatory-inhibitory network modelB: (c) Set spectral radius rsetsubscript𝑟setr_{\mathrm{set}}italic_r start_POSTSUBSCRIPT roman_set end_POSTSUBSCRIPT vsC: radii	BAC	BAC	CBA	BCA	Selection 3
A: The size of batches was set to 24 and 48 respectively for the two datasets.B: The size of batches is set to 128 and 512 respectively for the two datasets. cryoDRGN2 [33] was trained with its default parametersC: The ADAM optimizer with learning rate of 1.5×10−41.5superscript1041.5\times 10^{-4}1.5 × 10 start_POSTSUPERSCRIPT - 4 end_POSTSUPERSCRIPT was used to optimize our model	ABC	CBA	CAB	BAC	Selection 2
A: For example, the segmentation models have struggled to perform effectively on RGB images, such as bone marrow aspirate slides stained with Jenner-Giemsa. Furthermore, these models often require manual selection of both the model type and the specific image channel to be segmented, posing challenges for biologists with limited computational expertise.B: However, these algorithms were primarily trained using datasets consisting of gray-scale images and two-channel fluorescent images, lacking the necessary diversity to ensure robust generalization across a wide range of imaging modalitiesC: Efforts have been made towards the development of generalized cell segmentation algorithms [41, 8]	ACB	CBA	BCA	BCA	Selection 2
A: Again, for concreteness, we examine the first few terms of the right-hand sideB: We now use averaging theory to transform this non-autonomous system (25) into an autonomous systemC: The averaged order O⁢(ε)𝑂𝜀O(\varepsilon)italic_O ( italic_ε ) dynamics satisfy	BAC	CBA	CAB	BCA	Selection 1
A: Recently, the cases of non-Gaussian stable noises acting on QIFs started to attract the attention in mathematical neuroscienceB: In this paper, we explore the possibility of the implementation/generalization of the pseudocumulant approach for/to the populations of QIFs subject to δ𝛿\deltaitalic_δ-correlated non-Gaussian noise. C: However, currently, this interest is limited to the only exactly solvable case of a Cauchy noise. Pietras-etal-2023 ; Pyragas2-2023 The circular cumulant formalism was found useful for dealing with non-Gaussian noises, Dolmatova-Tyulkina-Goldobin-2023 but specifically for the case of QIFs the more recent formalism of pseudocumulants can be even more promising	CBA	ACB	BAC	BAC	Selection 2
A: The repository contains the Malmo environment code, training scripts for both the predictive coding and autoencoding neural networks, as well as code for the analysis of predictive coding and autoencoding resultsB: The code supporting the conclusions of this study is available on GitHub at https://github.com/jgornet/predictive-coding-recovers-mapsC: Should there be any questions or need for clarifications about the codebase, we encourage readers to raise an issue on the repository or reach out to the corresponding author.	ABC	CBA	BAC	CAB	Selection 3
A: (2018), to supplement the EEG encoder as "spatial filters" to encapsulate electrode correlations and reflect the spatial dynamics of brain activity.B: We utilize two approaches, self-attention (SA) Vaswani et alC: (2017), and graph attention (GA) Veličković et al	CBA	CAB	BAC	BCA	Selection 2
A: Such rooted trees exist when the number of species is as few as four. Similarly, the democratic vote estimate can be positively misleading for some unrooted trees with as few as five species. Species trees are in the anomaly zone if the gene tree with maximum probability is discordant from the species tree (Degnan and Rosenberg (2006)). However, by using supertree methods such as ASTRAL (Mirarab et alB: If the gene trees are assumed independent of each other, then the “democratic vote” estimator finds the species tree with the highest probability by counting the number of times each branching pattern appears in the list of gene trees. As more independent gene trees are accumulated, the gene tree with the highest probability obtains the most votes almost surely. As in Degnan and Rosenberg (2006), this estimate of the species tree is simply the gene tree topology that occurs most often. Even under these ideal assumptions, there exist species trees for which the democratic vote of MSC gene trees is positively misleadingC: (2014)) on unrooted quartets, any unrooted species tree can be consistently estimated in a polynomial number of species and polynomial number of gene trees. The ASTRAL suite has found extensive usage across many biological datasets. Finite sample guarantees have also been developed, see Shekhar et al. (2018). This type of result assumes some level of error tolerance ϵitalic-ϵ\epsilonitalic_ϵ, then provides a minimum number of genes that are required to obtain a provable amount of error below the tolerance level.	BAC	ABC	CBA	ABC	Selection 1
A: We also note the work of Weiss [42], who determined first-hitting-time estimates when the walkers are initially uniformly distributed throughout the domain.B: Relevant to our paper, they argue that the relative benefit of increased numbers of walkers is relatively weak (scaling as 1/log⁡(N)1𝑁1/\log(N)1 / roman_log ( italic_N )), and so it remains unclear if many biological systems are in this regimeC: In fact, Lawley and Madrid [24] have extended these works to also consider the effect of small noise	CBA	BAC	BCA	BAC	Selection 1
A: Furthermore, the relationship between linear transformers allowing for faster autoregressive learning and RNNs was established in a recent work [47]. Taken together, our current work would be a starting point to establish the bridge between the biological learning (towards the science of specialized brain circuits) and transformer learning within the seminal predictive coding hypothesis, which can be put in the theoretically solid variational free energy minimization conceptual framework.B: In addition, gated recurrent neural networks with multiplicative mechanisms were recently shown to be able to learn to implement linear self-attention [46]C: A recent work demonstrated that the weight uncertainty with the form of SaS structure can be also incorporated into the transformer [45]	CBA	BAC	ABC	BAC	Selection 1
A: Furthermore, our results are in agreement with the manually annotated results, highlighting the capability of our proposed toolbox in facilitating biological studies. The consistency between our automated analysis and the manually derived findings demonstrates the reliability and effectiveness of our approach in cellular analysis, offering valuable insights for further research and experimentation.B: Our results indicate that in the later stage of organoid formation (day 18), a higher concerntration of Geltrex leads to smaller organoid sizes, which aligns with the hypothesis that Geltrex, being a hydrogel, undergoes solidification at 37 degrees Celsius, thereby exerting pressure on organoid formation from the paper [5]C: We also analyzed the morphological features of organoids among different groups in Fig, 4	BCA	BAC	CBA	ABC	Selection 3
A: Such a mechanism does not seem to be biologically plausible for BNNs, as many authors have pointed out. Parameter update in BNNs occurs only locally, and distant neurons are only indirectly connected through the endogenous reward system. This observation is closely related to the weight transportation problem [6, 2, 4]. We refer to [12, 11] for a detailed discussion about the role of SGD in BNN, which the author of [10, Section 5] summarizes as follows:B: In simple terms, an ANN learns from data by adjusting the weights of the connections between nodes in order to minimize a loss function that measures the difference between the desired output and the actual output of the network. More specifically, the optimization step is performed using the Stochastic Gradient Descent (SGD) algorithm, which iteratively updates the weights of the network by moving them in the direction of the steepest descent of the empirical loss function of a single training sampleC: The gradient itself is computed with the so-called backpropagation algorithm. In particular, the update of any parameter is based on the states of all other parameters	CBA	CAB	ACB	ABC	Selection 2
A: During training, we repeated this process 100 times for each of the numbers 0-9, achieving perfect classification of all 10 digits with noise-free inference measurements. If we also take the noise into account during inference, we still achieve an overall accuracy of 95%, highlighting the system’s robustness against noise. Note that actual training is only performed on a simple and small neural network, that would otherwise not be capable of handling temporal inputs, while the “hard” work of separating the time-dependent signals is handled by the internal physics of our fluidic memristor. Ultimately, this successful classification of simple digit images serves as an explanatory proof-of-concept for the broader application of performing complex time-dependent data analysis tasks. B: This protocol is schematically illustrated in Fig. 4(c). Other types of simple readout functions could possibly also suffice. We trained our read-out network in silico using the results shown in Fig. 4(a, bottom). To incorporate the (device-to-device) variability, each individual pulse was subject to some noise newly drawn from a normal distribution with mean 0 and standard deviation given by the experimentally determined standard deviation for that specific voltage trainC: To illustrate how the results shown in Fig. 4(a) can be leveraged to classify more complex data inputs with an explanatory example, let us consider the simple single-digit numbers 0-9, represented by black and white 4×5454\times 54 × 5 pixel images. By converting a row of 4 pixels to a string of bits by letting a white pixel correspond to a “0” and a black pixel to a “1”, we can encode the entire image with 5 strings of 4 bits, as shown in Fig. 4(b) for the number “2” (other digits are shown in the SI). These bit-strings then generate 5 distinct signature outputs, as we saw in Fig. 4(a). A single-layer fully connected 5×105105\times 105 × 10 neural network is then trained in silico to classify the 5 measured conductances as numbers	CAB	CBA	ACB	ABC	Selection 2
A: In endemic zones, the sexual transmission route can substantially worsen the vulnerability of both mother and fetus to other sexually transmitted infections, particularly HIV rothan2018 .B: Both viruses can be transmitted through sexual contact and vertically from mother to fetusC: Although HIV is not a zoonotic disease, it has similar specific transmission mechanisms with ZIKV	BCA	CBA	ACB	ACB	Selection 2
A: All the other synaptic currents into the STN cells and the GP cells become downscaled due to DA effects. More details on the DA effects on the SPNs and synaptic currents are given in Appendices A and B, respectively.B: Such changes occur due to different DA effects, depending on the D1 and D2 SPNs. D1 receptor activation has two opposing effects. Due to a hyperpolarizing effect, activation threshold is increased in comparison to the bare case, while after threshold, the slope of the f-I curve increases rapidly because of another depolarizing effectC: In contrast, in the case of D2 SPN, only the depolarizing effect occurs, leading to left-shift of the bare f-I curve. As a result of DA effects, excitatory cortical inputs into the D1 (D2) SPNs are upscaled (downscaled), as shown well in Fig. 2C of Ref. SPN1	ACB	ACB	ACB	CAB	Selection 4
A: A limitation is that currently, there are relatively few model classes for which the Rashomon set can be computedB: Therefore, future work should aim to compute and store the Rashomon set of a wider variety of model classes. Future work may investigate incorporating Rashomon sets that may be well-approximated (e.g., GAMs, [10]), but not computed exactly, into the RID approach. Nonetheless, sparse trees are highly flexible, and using them with RID improves the trustworthiness and transparency of variable importance measures, enabling researchers to uncover important, reproducible relationships about complex processes without being misled by the Rashomon effect. C: RID can be directly computed for any model class for which the Rashomon set can be found – at the time of publishing, decision trees, linear models, and GLMs	CAB	BCA	ACB	CAB	Selection 2

**A**: In total in this dataset there are 220,177 images of 50 x 50 pixels in three colors**B**: The ratio between the two classes is 70 % normal and 30 % tumor for this dataset (Figure 2). **C**: Images classified as tumor by a pathologist are labelled as 1 while normal tissue images are labelled 0

ACB

CAB

BCA

CBA

Selection 1

**A**: We could however allow for more general initial conditions, as long as they lead to an asymptotic ESC within finitely many steps of the ln⁡K𝐾\ln Kroman_ln italic_K-algorithm. **B**: Both results assume that the system starts out in an asymptotic ESC**C**: These are the natural initial conditions, particularly when a first transition between asymptotic ESCs has already occurred

CBA

BCA

BAC

CAB

Selection 4

**A**: To date, a multitude of mathematical models describing infectious diseases through differential equations have been formulated and scrutinized to understand the dynamics of infection spread, exemplified by research on [1, 2, 3, 4]**B**: Recently, the mathematical modeling of the COVID-19 pandemic has captivated the attention of numerous experts, including mathematicians, scientists, epidemiologists, pharmacists, and chemists**C**: The outcomes of these endeavors have yielded several noteworthy and crucial results, as highlighted in the works of [4, 5, 6, 7, 8]

CBA

ABC

ACB

BAC

Selection 2

**A**: (2010) reported a heritability of 0.42 for default-mode network (DMN) in an extended**B**: Glahn et al**C**: One of the earliest papers on functional brain activation in twins is based on the resting-state EEG (Lykken et al., 1982), where they observed high twin correlation in MZ-twins on EEG spectra

BCA

ABC

CBA

BCA

Selection 3

**A**: Therefore, in this note we do not differentiate between the two cell phenotypes and focus instead on therapy effects. We consider tumor cells interacting with their environment, whose main physical components are the anisotropic brain tissue as well as brain-own and tumor-generated vasculature. We also modeled explicitly the evolution of VEGF controlling angiogenesis and the necrotic component of the neoplasm, which is relevant for assessing the tumor stage and for the segmentation needed for treatment planning**B**: The model investigates the impact of including microscopic dynamics also in the equation for endothelial cells, which has not been done before, comparing our approach with other two possible descriptions of EC dynamics. Then, we analyze via simulations various therapy approaches and their effects on tumor development and on normal tissue, thereby following treatment schedules addressed in clinical studies available in the literature. Although this has been done in [26] in a rather rudimentary way, we do not include tumor resection as part of the therapy; the problems have already been stated in that reference: it is hardly possible to provide a proper mathematical characterization of the tissue and neoplasm dynamics in the resected region in a continuous manner and the solution infers sharp discontinuities, which are difficult to handle. Considering the real data employed in Subsection 4.3, the simulations showed qualitatively reasonable results, however, a quantitative assessment could not be performed, due to missing data, as that patient’s therapy was stopped. **C**: The model proposed in this note is a therapy-oriented development of that considered in [7] and aligns to the approaches in [6, 8, 9, 13, 14, 15, 26]. Unlike [7, 15] we do not differentiate here between moving and proliferating cancer cells, but account for a single population of cells forming the tumor. In fact, the analysis in [7] showed (at least in our framework) that the main effect of splitting the tumor population into moving and resting (and proliferating) cells was the slower evolution of tumor mass, but qualitatively it behaved similarly to the situation with one population of moving and proliferating cancer cells

ABC

CBA

BCA

CAB

Selection 3

**A**: The reduction in the number of sulci in the brains of manatees suggests evolutionary trade-offs in brain gyrification**B**: For species with low risks of head impact—such as manatees—a lissencephalic structure, devoid of sulci, might facilitate more direct and efficient neural pathways. Conversely, species like humans, who must balance the risks of head impact with the need for neural efficiency, may benefit from moderate gyrification. **C**: One possible trade-off could be that sulci may compromise efficient neural connectivity, as illustrated in Fig. 11

ACB

CBA

BCA

CBA

Selection 1

**A**: Subsequently, synergistic and antagonistic classes (i.e. labels) were defined for each synergy score using thresholding based on the SynergyFinder software’s documentation333https://synergyfinder.fimm.fi/synergy/synfin_docs/#datanal, applying recommended thresholds (≥10absent10\geq 10≥ 10 for synergy and ≤−10absent10\leq-10≤ - 10 for antagonism) individually to the four synergy scores. **B**: The initial dataset underwent the following steps: Triplets with corresponding identifiers in the gene expression features table were selected**C**: Samples with missing values in drug or cell line identifiers, absent synergy scores, or duplicated triplets were filtered out

BCA

CBA

CAB

BAC

Selection 3

**A**: 2B).**B**: 2F), there is an optimal number of chromosomes for evolving novel traits (Fig**C**: We found that the skewness of the distribution of genotypes coded in each chromosome is essential to determining the probability of evolving new traits. Because skewness peaked with a finite number of chromosomes (Fig

BAC

CBA

BAC

BCA

Selection 2

**A**: The curl components are zero for edges that are not a 2-simplex boundary and the entries sum to zero around each node. The harmonic component sums to zero around each node, and it also sums to zero along each 2-simplex. We tested the topological equivalence of female brain networks and male brain networks using the Wasserstein distance (5).**B**: We further determined if we can detect topological differences in the decomposed components (Figure 2)**C**: The gradient component sum to zero along any cycles

CAB

BCA

CBA

ACB

Selection 1

**A**: in the case of populations suffering recurrent bottlenecks [CasanovaMiroJegousse2020] or overlapping generations (Casanova et al**B**: Others are modifications, such as models with varying population sizes [Mohle2002, KajKrone2003, Freund2020] or diploid reproduction [MohleSagitov2003, BirknerLiuSturm2018], using heuristics close to those of [MohleSagitov2001]. Finally some works use different techniques such as duality, e.g**C**: preprint arXiv:2210.05819).

BCA

ACB

BAC

ACB

Selection 3

**A**: To assess the effectiveness of the proposed model in capturing the morphological features of prostate cancer, a random subset of the WSIs was selected from the test set. Heatmaps, as shown in Fig**B**: RGB color maps were used to enhance the clarity, with red representing high attention and blue indicating low attention. These heat maps were overlaid onto the original WSIs. **C**: 4, were generated using a patch size of 224×224224224224\times 224224 × 224 and 90% overlay. The attention score of each patch was normalized to the range [0, 1], with high scores indicating regions crucial for diagnosis

BCA

ACB

BCA

Selection 3

**A**: The dataset was generated by filtering examples from the Protein Data Bank (PDB) corresponding to the same protein but bound to different biomolecules, with additional quality control criteria. For the scope of this work, we only focus on protein pairs where the provided Root Mean Square Deviation (RMSD) of the Cα𝛼\alphaitalic_α carbon atoms between unbound and bound 3D structures is >3.0absent3.0>3.0> 3.0Å, which amounts to 2370 examples in the D3PM dataset. **B**: Here we utilize the recently proposed D3PM dataset (Peng et al., 2022) that provides protein structures before (apo) and after (holo) binding, covering various types of protein motions**C**: The task of modeling conformational changes starting from a given protein structure is largely unexplored, mainly due to the lack of high-quality large datasets

CAB

ABC

BCA

CBA

Selection 4

**A**: A characteristic equation was also deduced for the special case when neither the growth rate g𝑔gitalic_g nor the mortality rate μ𝜇\muitalic_μ depend on the interaction variable E𝐸Eitalic_E (β𝛽\betaitalic_β on the other hand does). Note however that the linearisation and stability results established in [15] were completely formal, as the Principle of Linearised Stability has not been established for the PDE formulation (1). This is the main reason why in the current work we employ a different formulation of the model. **B**: On the other hand, in [15] the authors derived a formal linearisation of the model and studied regularity properties of the governing linear semigroup**C**: We note that the focus in [1, 2] was on numerical approximation of solutions of the hierarchical model

ACB

CBA

BCA

Selection 2

**A**: The classical representation of coloured graphs is through their colour classes**B**: In this section we introduce a different representation of pdCGs that will allow us to deal more efficiently with these objects**C**: A detailed example that may help to follow the steps required for the construction of the novel representation is given in Section A of the Appendix.

CAB

BCA

ABC

CAB

Selection 3

**A**: We standardize cardiac MRI intensity levels using Z-score normalization [7] to eliminate inter-subject variations**B**: Furthermore, we detect automatic landmarks which is explained in the next paragraph. We perform affine registration to align the heart regions of different subjects to a target image space. We then carry out in-plane scaling of scans by max-pooling at 2222, 4444, 8888, and 16161616 times and obtain down-sampled resolutions of 128×128128128128\times 128128 × 128, 64×64646464\times 6464 × 64, 32×32323232\times 3232 × 32, and 16×16161616\times 1616 × 16, respectively.**C**: Cardiac MRI Preprocessing: The preprocessing of cardiac MRI contains (1111) normalization of scans, (2222) automatic landmark detection, (3333) inter-subject registration, and (4444) in-plane downsampling

CAB

ACB

BCA

CBA

Selection 3

**A**: This paper addresses two challenges posed by the multimodal fusion of transcriptomics and histology: (1) we address the challenge of transcriptomics tokenization by defining biological pathway tokens that encode semantically meaningful and interpretable functions, and (2) we overcome the computational challenge of integrating long multimodal sequences by designing a multimodal Transformer with sparse modality-specific attention patterns. Our model achieves state-of-the-art survival performance when tested on five datasets from TCGA**B**: In addition, our interpretability framework reveals known and candidate prognostic features. While our interpretability framework enables identifying prognostic features, these findings remain qualitative**C**: Future work could focus on interpretability metrics that generalize findings at dataset-level, e.g., with quantitative morphological characterizations of specific pathways. In addition, our findings suggest that including patch-to-patch interactions does not lead to improved performance. Nonetheless, the absence of a performance boost should not be an evidence that patch-to-patch interactions are unnecessary, but rather that modeling such interactions is a challenging problem that remains to be solved.

BCA

CAB

ABC

BCA

Selection 3

**A**: where we have applied Stirling’s approximation [30] to derive an asymptotic expression for the large compartment number, ν≫1much-greater-than𝜈1\nu\gg 1italic_ν ≫ 1, limit**B**: Even for ν=1𝜈1\nu=1italic_ν = 1, the asymptotic expression produces excellent agreement with simulation results.**C**: In fig. 2a, we compare both the exact and asymptotic expressions for the stationary variance to a numerical approximation produced through repeated simulation of the SDE

ABC

ACB

ABC

BCA

Selection 2

**A**: For each dataset, it establishes the preferred metric for evaluation to enable consistent comparison across models. We describe each dataset selected to evaluate our model.**B**: MoleculeNet is a benchmark set used to evaluate machine learning techniques [9]**C**: It curate’s quantum mechanics, physical chemistry, biophysics, and physiology datasets

CAB

ACB

ABC

ACB

Selection 1

**A**: In the quenched regime, the network changes very slowly compared to the evolution of the epidemic**B**: Indeed, as we will show in Section I.3, the quenched regime assumes that, before each topology update, the epidemic has almost reached its equilibrium. **C**: Quenched processes are well approximated with processes on static networks and therefore well studied over the last two decades

ABC

BAC

ACB

Selection 4

**A**: We allow degenerate vertices with in-degree 2222 as well as out-degree 2**B**: The cover 𝒞𝒞\mathcal{C}caligraphic_C for N𝑁Nitalic_N then consists of sets of size 1 or 2, and each integer appearing in 𝒞𝒞\mathcal{C}caligraphic_C appears either once, if it is a tree vertex (in-degree 1), or twice if it is a reticulation (in-degree 2). **C**: Suppose N𝑁Nitalic_N is a binary tree-based network

ABC

BCA

ABC

BAC

Selection 2

**A**: As pointed out by Barton et al. (2004), the Moran model with finite N𝑁Nitalic_N is reversible, meaning that at stationarity the time-reversed process is the same (in distribution) as the forward-time Moran process**B**: This is not a property of the Wright-Fisher model with finite N𝑁Nitalic_N but does hold for their shared diffusion limit (1) with stationary density (2); see for instance Millet et al. (1989) for why this holds. Figure 1 gives an illustration of a genealogy with mutations and allele frequencies varying backward in time.**C**: Even if the sample frequencies are known, the allele frequencies in the population are unknown. A key feature of this method is to model allele-frequency trajectories backward in time

CBA

BAC

BCA

ACB

Selection 3

**A**: We tackle this problem by assuming that each read is coming from the one (e.g**B**: ’maternal’) chromosome with probability w𝑤witalic_w and from the other chromosome (e.g**C**: paternal) with probability 1−w1𝑤1-w1 - italic_w, where the balance between w𝑤witalic_w and 1−w1𝑤1-w1 - italic_w reflects BAD. This is done naturally with the mixture distribution:

ABC

BAC

BCA

Selection 1

**A**: We set Ci,k+1=Isubscript𝐶𝑖𝑘1𝐼C_{i,k+1}=Iitalic_C start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_I and τi,k+1=τi,k+Zi,ksubscript𝜏𝑖𝑘1subscript𝜏𝑖𝑘subscript𝑍𝑖𝑘\tau_{i,k+1}=\tau_{i,k}+Z_{i,k}italic_τ start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_τ start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPT + italic_Z start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPT**B**: vaccinated**C**: Otherwise, the individual is vaccinated before being reinfected, and we set Ci,k+1=Ssubscript𝐶𝑖𝑘1𝑆C_{i,k+1}=Sitalic_C start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_S, and τi,k+1=τi,k+TV,i,ksubscript𝜏𝑖𝑘1subscript𝜏𝑖𝑘subscript𝑇𝑉𝑖𝑘\tau_{i,k+1}=\tau_{i,k}+T_{V,i,k}italic_τ start_POSTSUBSCRIPT italic_i , italic_k + 1 end_POSTSUBSCRIPT = italic_τ start_POSTSUBSCRIPT italic_i , italic_k end_POSTSUBSCRIPT + italic_T start_POSTSUBSCRIPT italic_V , italic_i , italic_k end_POSTSUBSCRIPT.

CAB

BCA

CBA

BAC

Selection 4

**A**: acknowledges Iniziativa PNC0000002-DARE - Digital Lifelong Prevention. S.A., F.F., and A.M. also acknowledge the support by the Italian Ministry of University and Research (project funded by the European Union - Next Generation EU: “PNRR Missione 4 Componente 2, “Dalla ricerca all’impresa”, Investimento 1.4, Progetto CN00000033”).**B**: Acknowledgements. We wish to acknowledge Jacopo Grilli and Davide Bernardi for critical reading of the manuscript and useful discussions. F.F**C**: thanks Matteo Guardiani and the Information Field Theory group at the Max-Planck Institute for Astrophysics for their hospitality and helpful comments. S.S

CBA

CAB

ACB

BCA

Selection 2

**A**: In the non-stationary case, the results are possible, but more involving**B**: The above results pertain to the stationary case**C**: The non-stationary queue length distribution can be computed using renewal theory provided one can compute the distribution of the forward recurrence time (the time until the next arrival) [72], whereas the moments can be computed recursively without the knowledge of this distribution [71].

BAC

CBA

CAB

Selection 1

**A**: The remainder of this paper consists of two sections**B**: The next section contains the proofs of our theoretical results from Section 2**C**: In Section 4 we provide a numerical example of an MTBDP where the rates depend on the number of particles of each type that are present in the system.

BAC

ABC

ACB

Selection 2

**A**: As a result, the RGCN encoder might struggle to capture complex patterns and may not fully leverage the underlying protein data, potentially leading to suboptimal performance.**B**: One limitation of our proposed Prot2Text model is that the RGCN encoder is not pretrained**C**: Unlike the ESM encoder, which benefits from pretraining on a large corpus, the RGCN encoder lacks this initial knowledge

ABC

CAB

CBA

BAC

Selection 2

**A**: measured spectral radius rsimsubscript𝑟simr_{\mathrm{sim}}italic_r start_POSTSUBSCRIPT roman_sim end_POSTSUBSCRIPT. Same excitatory-inhibitory network model**B**: (c) Set spectral radius rsetsubscript𝑟setr_{\mathrm{set}}italic_r start_POSTSUBSCRIPT roman_set end_POSTSUBSCRIPT vs**C**: radii

BAC

CBA

BCA

Selection 3

**A**: The size of batches was set to 24 and 48 respectively for the two datasets.**B**: The size of batches is set to 128 and 512 respectively for the two datasets. cryoDRGN2 [33] was trained with its default parameters**C**: The ADAM optimizer with learning rate of 1.5×10−41.5superscript1041.5\times 10^{-4}1.5 × 10 start_POSTSUPERSCRIPT - 4 end_POSTSUPERSCRIPT was used to optimize our model

ABC

CBA

CAB

BAC

Selection 2

**A**: For example, the segmentation models have struggled to perform effectively on RGB images, such as bone marrow aspirate slides stained with Jenner-Giemsa. Furthermore, these models often require manual selection of both the model type and the specific image channel to be segmented, posing challenges for biologists with limited computational expertise.**B**: However, these algorithms were primarily trained using datasets consisting of gray-scale images and two-channel fluorescent images, lacking the necessary diversity to ensure robust generalization across a wide range of imaging modalities**C**: Efforts have been made towards the development of generalized cell segmentation algorithms [41, 8]

ACB

CBA

BCA

Selection 2

**A**: Again, for concreteness, we examine the first few terms of the right-hand side**B**: We now use averaging theory to transform this non-autonomous system (25) into an autonomous system**C**: The averaged order O⁢(ε)𝑂𝜀O(\varepsilon)italic_O ( italic_ε ) dynamics satisfy

BAC

CBA

CAB

BCA

Selection 1

**A**: Recently, the cases of non-Gaussian stable noises acting on QIFs started to attract the attention in mathematical neuroscience**B**: In this paper, we explore the possibility of the implementation/generalization of the pseudocumulant approach for/to the populations of QIFs subject to δ𝛿\deltaitalic_δ-correlated non-Gaussian noise. **C**: However, currently, this interest is limited to the only exactly solvable case of a Cauchy noise. Pietras-etal-2023 ; Pyragas2-2023 The circular cumulant formalism was found useful for dealing with non-Gaussian noises, Dolmatova-Tyulkina-Goldobin-2023 but specifically for the case of QIFs the more recent formalism of pseudocumulants can be even more promising

CBA

ACB

BAC

Selection 2

**A**: The repository contains the Malmo environment code, training scripts for both the predictive coding and autoencoding neural networks, as well as code for the analysis of predictive coding and autoencoding results**B**: The code supporting the conclusions of this study is available on GitHub at https://github.com/jgornet/predictive-coding-recovers-maps**C**: Should there be any questions or need for clarifications about the codebase, we encourage readers to raise an issue on the repository or reach out to the corresponding author.

ABC

CBA

BAC

CAB

Selection 3

**A**: (2018), to supplement the EEG encoder as "spatial filters" to encapsulate electrode correlations and reflect the spatial dynamics of brain activity.**B**: We utilize two approaches, self-attention (SA) Vaswani et al**C**: (2017), and graph attention (GA) Veličković et al

CBA

CAB

BAC

BCA

Selection 2

**A**: Such rooted trees exist when the number of species is as few as four. Similarly, the democratic vote estimate can be positively misleading for some unrooted trees with as few as five species. Species trees are in the anomaly zone if the gene tree with maximum probability is discordant from the species tree (Degnan and Rosenberg (2006)). However, by using supertree methods such as ASTRAL (Mirarab et al**B**: If the gene trees are assumed independent of each other, then the “democratic vote” estimator finds the species tree with the highest probability by counting the number of times each branching pattern appears in the list of gene trees. As more independent gene trees are accumulated, the gene tree with the highest probability obtains the most votes almost surely. As in Degnan and Rosenberg (2006), this estimate of the species tree is simply the gene tree topology that occurs most often. Even under these ideal assumptions, there exist species trees for which the democratic vote of MSC gene trees is positively misleading**C**: (2014)) on unrooted quartets, any unrooted species tree can be consistently estimated in a polynomial number of species and polynomial number of gene trees. The ASTRAL suite has found extensive usage across many biological datasets. Finite sample guarantees have also been developed, see Shekhar et al. (2018). This type of result assumes some level of error tolerance ϵitalic-ϵ\epsilonitalic_ϵ, then provides a minimum number of genes that are required to obtain a provable amount of error below the tolerance level.

BAC

ABC

CBA

ABC

Selection 1

**A**: We also note the work of Weiss [42], who determined first-hitting-time estimates when the walkers are initially uniformly distributed throughout the domain.**B**: Relevant to our paper, they argue that the relative benefit of increased numbers of walkers is relatively weak (scaling as 1/log⁡(N)1𝑁1/\log(N)1 / roman_log ( italic_N )), and so it remains unclear if many biological systems are in this regime**C**: In fact, Lawley and Madrid [24] have extended these works to also consider the effect of small noise

CBA

BAC

BCA

BAC

Selection 1

**A**: Furthermore, the relationship between linear transformers allowing for faster autoregressive learning and RNNs was established in a recent work [47]. Taken together, our current work would be a starting point to establish the bridge between the biological learning (towards the science of specialized brain circuits) and transformer learning within the seminal predictive coding hypothesis, which can be put in the theoretically solid variational free energy minimization conceptual framework.**B**: In addition, gated recurrent neural networks with multiplicative mechanisms were recently shown to be able to learn to implement linear self-attention [46]**C**: A recent work demonstrated that the weight uncertainty with the form of SaS structure can be also incorporated into the transformer [45]

CBA

BAC

ABC

BAC

Selection 1

**A**: Furthermore, our results are in agreement with the manually annotated results, highlighting the capability of our proposed toolbox in facilitating biological studies. The consistency between our automated analysis and the manually derived findings demonstrates the reliability and effectiveness of our approach in cellular analysis, offering valuable insights for further research and experimentation.**B**: Our results indicate that in the later stage of organoid formation (day 18), a higher concerntration of Geltrex leads to smaller organoid sizes, which aligns with the hypothesis that Geltrex, being a hydrogel, undergoes solidification at 37 degrees Celsius, thereby exerting pressure on organoid formation from the paper [5]**C**: We also analyzed the morphological features of organoids among different groups in Fig, 4

BCA

BAC

CBA

ABC

Selection 3

**A**: Such a mechanism does not seem to be biologically plausible for BNNs, as many authors have pointed out. Parameter update in BNNs occurs only locally, and distant neurons are only indirectly connected through the endogenous reward system. This observation is closely related to the weight transportation problem [6, 2, 4]. We refer to [12, 11] for a detailed discussion about the role of SGD in BNN, which the author of [10, Section 5] summarizes as follows:**B**: In simple terms, an ANN learns from data by adjusting the weights of the connections between nodes in order to minimize a loss function that measures the difference between the desired output and the actual output of the network. More specifically, the optimization step is performed using the Stochastic Gradient Descent (SGD) algorithm, which iteratively updates the weights of the network by moving them in the direction of the steepest descent of the empirical loss function of a single training sample**C**: The gradient itself is computed with the so-called backpropagation algorithm. In particular, the update of any parameter is based on the states of all other parameters

CBA

CAB

ACB

ABC

Selection 2

**A**: During training, we repeated this process 100 times for each of the numbers 0-9, achieving perfect classification of all 10 digits with noise-free inference measurements. If we also take the noise into account during inference, we still achieve an overall accuracy of 95%, highlighting the system’s robustness against noise. Note that actual training is only performed on a simple and small neural network, that would otherwise not be capable of handling temporal inputs, while the “hard” work of separating the time-dependent signals is handled by the internal physics of our fluidic memristor. Ultimately, this successful classification of simple digit images serves as an explanatory proof-of-concept for the broader application of performing complex time-dependent data analysis tasks. **B**: This protocol is schematically illustrated in Fig. 4(c). Other types of simple readout functions could possibly also suffice. We trained our read-out network in silico using the results shown in Fig. 4(a, bottom). To incorporate the (device-to-device) variability, each individual pulse was subject to some noise newly drawn from a normal distribution with mean 0 and standard deviation given by the experimentally determined standard deviation for that specific voltage train**C**: To illustrate how the results shown in Fig. 4(a) can be leveraged to classify more complex data inputs with an explanatory example, let us consider the simple single-digit numbers 0-9, represented by black and white 4×5454\times 54 × 5 pixel images. By converting a row of 4 pixels to a string of bits by letting a white pixel correspond to a “0” and a black pixel to a “1”, we can encode the entire image with 5 strings of 4 bits, as shown in Fig. 4(b) for the number “2” (other digits are shown in the SI). These bit-strings then generate 5 distinct signature outputs, as we saw in Fig. 4(a). A single-layer fully connected 5×105105\times 105 × 10 neural network is then trained in silico to classify the 5 measured conductances as numbers

CAB

CBA

ACB

ABC

Selection 2

**A**: In endemic zones, the sexual transmission route can substantially worsen the vulnerability of both mother and fetus to other sexually transmitted infections, particularly HIV rothan2018 .**B**: Both viruses can be transmitted through sexual contact and vertically from mother to fetus**C**: Although HIV is not a zoonotic disease, it has similar specific transmission mechanisms with ZIKV

BCA

CBA

ACB

Selection 2

**A**: All the other synaptic currents into the STN cells and the GP cells become downscaled due to DA effects. More details on the DA effects on the SPNs and synaptic currents are given in Appendices A and B, respectively.**B**: Such changes occur due to different DA effects, depending on the D1 and D2 SPNs. D1 receptor activation has two opposing effects. Due to a hyperpolarizing effect, activation threshold is increased in comparison to the bare case, while after threshold, the slope of the f-I curve increases rapidly because of another depolarizing effect**C**: In contrast, in the case of D2 SPN, only the depolarizing effect occurs, leading to left-shift of the bare f-I curve. As a result of DA effects, excitatory cortical inputs into the D1 (D2) SPNs are upscaled (downscaled), as shown well in Fig. 2C of Ref. SPN1

ACB

CAB

Selection 4

**A**: A limitation is that currently, there are relatively few model classes for which the Rashomon set can be computed**B**: Therefore, future work should aim to compute and store the Rashomon set of a wider variety of model classes. Future work may investigate incorporating Rashomon sets that may be well-approximated (e.g., GAMs, [10]), but not computed exactly, into the RID approach. Nonetheless, sparse trees are highly flexible, and using them with RID improves the trustworthiness and transparency of variable importance measures, enabling researchers to uncover important, reproducible relationships about complex processes without being misled by the Rashomon effect. **C**: RID can be directly computed for any model class for which the Rashomon set can be found – at the time of publishing, decision trees, linear models, and GLMs

CAB

BCA

ACB

CAB

Selection 2

Datasets:

liangzid
/

robench2024b_all_setq-bioSCP-s-50

Collection including liangzid/robench2024b_all_setq-bioSCP-s-50

robench-2024b