Jonathan Rubin

J. Ben Tamo, Daniel Carlander-Reuterfelt, Jonathan Rubin et al.

Despite multilingual pretraining, large language models often struggle with non-English tasks, particularly in language control, the ability to respond in the intended language. We identify and characterize two key failure modes: the multilingual transfer bottleneck (correct language, incorrect task response) and the language consistency bottleneck (correct task response, wrong language). To systematically surface these issues, we design a four-scenario evaluation protocol spanning MMLU, MGSM, and XQuAD benchmarks. To probe these issues with interpretability, we extend logit lens analysis to track language probabilities layer by layer and compute cross-lingual semantic similarity of hidden states. The results reveal a three-phase internal structure: early layers align inputs into a shared semantic space, middle layers perform task reasoning, and late layers drive language-specific generation. Guided by these insights, we introduce selective fine-tuning of only the final layers responsible for language control. On Qwen-3-32B and Bloom-7.1B, this method achieves over 98 percent language consistency across six languages while fine-tuning only 3-5 percent of parameters, without sacrificing task accuracy. Importantly, this result is nearly identical to that of full-scope fine-tuning (for example, above 98 percent language consistency for both methods across all prompt scenarios) but uses a fraction of the computational resources. To the best of our knowledge, this is the first approach to leverage layer-localization of language control for efficient multilingual adaptation.

Annamalai Natarajan, Gregory Boverman, Yale Chang et al.

We present our entry to the 2021 PhysioNet/CinC challenge - a waveform transformer model to detect cardiac abnormalities from ECG recordings. We compare the performance of the waveform transformer model on different ECG-lead subsets using approximately 88,000 ECG recordings from six datasets. In the official rankings, team prna ranked between 9 and 15 on 12, 6, 4, 3 and 2-lead sets respectively. Our waveform transformer model achieved an average challenge metric of 0.47 on the held-out test set across all ECG-lead subsets. Our combined performance across all leads placed us at rank 11 out of 39 officially ranking teams.

Asif Rahman, Yale Chang, Jonathan Rubin

Time series models with recurrent neural networks (RNNs) can have high accuracy but are unfortunately difficult to interpret as a result of feature-interactions, temporal-interactions, and non-linear transformations. Interpretability is important in domains like healthcare where constructing models that provide insight into the relationships they have learned are required to validate and trust model predictions. We want accurate time series models where users can understand the contribution of individual input features. We present the Interpretable-RNN (I-RNN) that balances model complexity and accuracy by forcing the relationship between variables in the model to be additive. Interactions are restricted between hidden states of the RNN and additively combined at the final step. I-RNN specifically captures the unique characteristics of clinical time series, which are unevenly sampled in time, asynchronously acquired, and have missing data. Importantly, the hidden state activations represent feature coefficients that correlate with the prediction target and can be visualized as risk curves that capture the global relationship between individual input features and the outcome. We evaluate the I-RNN model on the Physionet 2012 Challenge dataset to predict in-hospital mortality, and on a real-world clinical decision support task: predicting hemodynamic interventions in the intensive care unit. I-RNN provides explanations in the form of global and local feature importances comparable to highly intelligible models like decision trees trained on hand-engineered features while significantly outperforming them. I-RNN remains intelligible while providing accuracy comparable to state-of-the-art decay-based and interpolation-based recurrent time series models. The experimental results on real-world clinical datasets refute the myth that there is a tradeoff between accuracy and interpretability.

Ruizhi Liao, Jonathan Rubin, Grace Lam et al.

We propose and demonstrate machine learning algorithms to assess the severity of pulmonary edema in chest x-ray images of congestive heart failure patients. Accurate assessment of pulmonary edema in heart failure is critical when making treatment and disposition decisions. Our work is grounded in a large-scale clinical dataset of over 300,000 x-ray images with associated radiology reports. While edema severity labels can be extracted unambiguously from a small fraction of the radiology reports, accurate annotation is challenging in most cases. To take advantage of the unlabeled images, we develop a Bayesian model that includes a variational auto-encoder for learning a latent representation from the entire image set trained jointly with a regressor that employs this representation for predicting pulmonary edema severity. Our experimental results suggest that modeling the distribution of images jointly with the limited labels improves the accuracy of pulmonary edema scoring compared to a strictly supervised approach. To the best of our knowledge, this is the first attempt to employ machine learning algorithms to automatically and quantitatively assess the severity of pulmonary edema in chest x-ray images.

S. Mazdak Abulnaga, Jonathan Rubin

We present a fully convolutional neural network for segmenting ischemic stroke lesions in CT perfusion images for the ISLES 2018 challenge. Treatment of stroke is time sensitive and current standards for lesion identification require manual segmentation, a time consuming and challenging process. Automatic segmentation methods present the possibility of accurately identifying lesions and improving treatment planning. Our model is based on the PSPNet, a network architecture that makes use of pyramid pooling to provide global and local contextual information. To learn the varying shapes of the lesions, we train our network using focal loss, a loss function designed for the network to focus on learning the more difficult samples. We compare our model to networks trained using the U-Net and V-Net architectures. Our approach demonstrates effective performance in lesion segmentation and ranked among the top performers at the challenge conclusion.

Saman Parvaneh, Jonathan Rubin, Ali Samadani et al.

The visual scoring of arousals during sleep routinely conducted by sleep experts is a challenging task warranting an automatic approach. This paper presents an algorithm for automatic detection of arousals during sleep. Using the Physionet/CinC Challenge dataset, an 80-20% subject-level split was performed to create in-house training and test sets, respectively. The data for each subject in the training set was split to 30-second epochs with no overlap. A total of 428 features from EEG, EMG, EOG, airflow, and SaO2 in each epoch were extracted and used for creating subject-specific models based on an ensemble of bagged classification trees, resulting in 943 models. For marking arousal and non-arousal regions in the test set, the data in the test set was split to 30-second epochs with 50% overlaps. The average of arousal probabilities from different patient-specific models was assigned to each 30-second epoch and then a sample-wise probability vector with the same length as test data was created for model evaluation. Using the PhysioNet/CinC Challenge 2018 scoring criteria, AUPRCs of 0.25 and 0.21 were achieved for the in-house test and blind test sets, respectively.

Jonathan Rubin, Deepan Sanghavi, Claire Zhao et al.

The MIMIC-CXR dataset is (to date) the largest released chest x-ray dataset consisting of 473,064 chest x-rays and 206,574 radiology reports collected from 63,478 patients. We present the results of training and evaluating a collection of deep convolutional neural networks on this dataset to recognize multiple common thorax diseases. To the best of our knowledge, this is the first work that trains CNNs for this task on such a large collection of chest x-ray images, which is over four times the size of the largest previously released chest x-ray corpus (ChestX-Ray14). We describe and evaluate individual CNN models trained on frontal and lateral CXR view types. In addition, we present a novel DualNet architecture that emulates routine clinical practice by simultaneously processing both frontal and lateral CXR images obtained from a radiological exam. Our DualNet architecture shows improved performance in recognizing findings in CXR images when compared to applying separate baseline frontal and lateral classifiers.

Jonathan Rubin, Saman Parvaneh, Asif Rahman et al.

The development of new technology such as wearables that record high-quality single channel ECG, provides an opportunity for ECG screening in a larger population, especially for atrial fibrillation screening. The main goal of this study is to develop an automatic classification algorithm for normal sinus rhythm (NSR), atrial fibrillation (AF), other rhythms (O), and noise from a single channel short ECG segment (9-60 seconds). For this purpose, signal quality index (SQI) along with dense convolutional neural networks was used. Two convolutional neural network (CNN) models (main model that accepts 15 seconds ECG and secondary model that processes 9 seconds shorter ECG) were trained using the training data set. If the recording is determined to be of low quality by SQI, it is immediately classified as noisy. Otherwise, it is transformed to a time-frequency representation and classified with the CNN as NSR, AF, O, or noise. At the final step, a feature-based post-processing algorithm classifies the rhythm as either NSR or O in case the CNN model's discrimination between the two is indeterminate. The best result achieved at the official phase of the PhysioNet/CinC challenge on the blind test set was 0.80 (F1 for NSR, AF, and O were 0.90, 0.80, and 0.70, respectively).

Jonathan Rubin, Rui Abreu, Anurag Ganguli et al.

The work presented here applies deep learning to the task of automated cardiac auscultation, i.e. recognizing abnormalities in heart sounds. We describe an automated heart sound classification algorithm that combines the use of time-frequency heat map representations with a deep convolutional neural network (CNN). Given the cost-sensitive nature of misclassification, our CNN architecture is trained using a modified loss function that directly optimizes the trade-off between sensitivity and specificity. We evaluated our algorithm at the 2016 PhysioNet Computing in Cardiology challenge where the objective was to accurately classify normal and abnormal heart sounds from single, short, potentially noisy recordings. Our entry to the challenge achieved a final specificity of 0.95, sensitivity of 0.73 and overall score of 0.84. We achieved the greatest specificity score out of all challenge entries and, using just a single CNN, our algorithm differed in overall score by only 0.02 compared to the top place finisher, which used an ensemble approach.