Nikhil Garg

J. D. Zamfirescu-Pereira, Jerry Chen, Emily Wen et al. · berkeley

Algorithms provide powerful tools for detecting and dissecting human bias and error. Here, we develop machine learning methods to to analyze how humans err in a particular high-stakes task: image interpretation. We leverage a unique dataset of 16,135,392 human predictions of whether a neighborhood voted for Donald Trump or Joe Biden in the 2020 US election, based on a Google Street View image. We show that by training a machine learning estimator of the Bayes optimal decision for each image, we can provide an actionable decomposition of human error into bias, variance, and noise terms, and further identify specific features (like pickup trucks) which lead humans astray. Our methods can be applied to ensure that human-in-the-loop decision-making is accurate and fair and are also applicable to black-box algorithmic systems.

Rajiv Movva, Divya Shanmugam, Kaihua Hou et al.

Healthcare data in the United States often records only a patient's coarse race group: for example, both Indian and Chinese patients are typically coded as "Asian." It is unknown, however, whether this coarse coding conceals meaningful disparities in the performance of clinical risk scores across granular race groups. Here we show that it does. Using data from 418K emergency department visits, we assess clinical risk score performance disparities across 26 granular groups for three outcomes, five risk scores, and four performance metrics. Across outcomes and metrics, we show that the risk scores exhibit significant granular performance disparities within coarse race groups. In fact, variation in performance within coarse groups often *exceeds* the variation between coarse groups. We explore why these disparities arise, finding that outcome rates, feature distributions, and the relationships between features and outcomes all vary significantly across granular groups. Our results suggest that healthcare providers, hospital systems, and machine learning researchers should strive to collect, release, and use granular race data in place of coarse race data, and that existing analyses may significantly underestimate racial disparities in performance.

Nikhil Garg, Ismael Balafrej, Terrence C. Stewart et al.

This study proposes voltage-dependent-synaptic plasticity (VDSP), a novel brain-inspired unsupervised local learning rule for the online implementation of Hebb's plasticity mechanism on neuromorphic hardware. The proposed VDSP learning rule updates the synaptic conductance on the spike of the postsynaptic neuron only, which reduces by a factor of two the number of updates with respect to standard spike-timing-dependent plasticity (STDP). This update is dependent on the membrane potential of the presynaptic neuron, which is readily available as part of neuron implementation and hence does not require additional memory for storage. Moreover, the update is also regularized on synaptic weight and prevents explosion or vanishing of weights on repeated stimulation. Rigorous mathematical analysis is performed to draw an equivalence between VDSP and STDP. To validate the system-level performance of VDSP, we train a single-layer spiking neural network (SNN) for the recognition of handwritten digits. We report 85.01 $ \pm $ 0.76% (Mean $ \pm $ S.D.) accuracy for a network of 100 output neurons on the MNIST dataset. The performance improves when scaling the network size (89.93 $ \pm $ 0.41% for 400 output neurons, 90.56 $ \pm $ 0.27 for 500 neurons), which validates the applicability of the proposed learning rule for spatial pattern recognition tasks. Future work will consider more complicated tasks. Interestingly, the learning rule better adapts than STDP to the frequency of input signal and does not require hand-tuning of hyperparameters

Nikhil Garg, Hannah Li, Faidra Monachou

We study the role of information and access in capacity-constrained selection problems with fairness concerns. We develop a statistical discrimination framework, where each applicant has multiple features and is potentially strategic. The model formalizes the trade-off between the (potentially positive) informational role of a feature and its (negative) exclusionary nature when members of different social groups have unequal access to this feature. Our framework finds a natural application to policy debates on dropping standardized testing in admissions. Our primary takeaway is that the decision to drop a feature (such as test scores) cannot be made without the joint context of the information provided by other features and how the requirement affects the applicant pool composition. Dropping a feature may exacerbate disparities by decreasing the amount of information available for each applicant, especially those from non-traditional backgrounds. However, in the presence of access barriers to a feature, the interaction between the informational environment and the effect of access barriers on the applicant pool size becomes highly complex. Furthermore, we consider an extension with two schools and costly tests, where strategic students decide whether to take the test or not. Our theoretical results reveal that the students' test-taking behavior can be non-monotonic. We characterize the two-school policy equilibria and show that each school's optimal decision to drop the test critically depends on the other school's test policy. Finally, using calibrated simulations, we demonstrate the presence of practical instances where the decision to eliminate standardized testing improves or worsens all metrics.

Rajiv Movva, Sidhika Balachandar, Kenny Peng et al.

Large language models (LLMs) are dramatically influencing AI research, spurring discussions on what has changed so far and how to shape the field's future. To clarify such questions, we analyze a new dataset of 16,979 LLM-related arXiv papers, focusing on recent trends in 2023 vs. 2018-2022. First, we study disciplinary shifts: LLM research increasingly considers societal impacts, evidenced by 20x growth in LLM submissions to the Computers and Society sub-arXiv. An influx of new authors -- half of all first authors in 2023 -- are entering from non-NLP fields of CS, driving disciplinary expansion. Second, we study industry and academic publishing trends. Surprisingly, industry accounts for a smaller publication share in 2023, largely due to reduced output from Google and other Big Tech companies; universities in Asia are publishing more. Third, we study institutional collaboration: while industry-academic collaborations are common, they tend to focus on the same topics that industry focuses on rather than bridging differences. The most prolific institutions are all US- or China-based, but there is very little cross-country collaboration. We discuss implications around (1) how to support the influx of new authors, (2) how industry trends may affect academics, and (3) possible effects of (the lack of) collaboration.

Sidhika Balachandar, Nikhil Garg, Emma Pierson

Machine learning models are often trained to predict the outcome resulting from a human decision. For example, if a doctor decides to test a patient for disease, will the patient test positive? A challenge is that historical decision-making determines whether the outcome is observed: we only observe test outcomes for patients doctors historically tested. Untested patients, for whom outcomes are unobserved, may differ from tested patients along observed and unobserved dimensions. We propose a Bayesian model class which captures this setting. The purpose of the model is to accurately estimate risk for both tested and untested patients. Estimating this model is challenging due to the wide range of possibilities for untested patients. To address this, we propose two domain constraints which are plausible in health settings: a prevalence constraint, where the overall disease prevalence is known, and an expertise constraint, where the human decision-maker deviates from purely risk-based decision-making only along a constrained feature set. We show theoretically and on synthetic data that domain constraints improve parameter inference. We apply our model to a case study of cancer risk prediction, showing that the model's inferred risk predicts cancer diagnoses, its inferred testing policy captures known public health policies, and it can identify suboptimalities in test allocation. Though our case study is in healthcare, our analysis reveals a general class of domain constraints which can improve model estimation in many settings.

Erica Chiang, Divya Shanmugam, Ashley N. Beecy et al.

Disease progression models are widely used to inform the diagnosis and treatment of many progressive diseases. However, a significant limitation of existing models is that they do not account for health disparities that can bias the observed data. To address this, we develop an interpretable Bayesian disease progression model that captures three key health disparities: certain patient populations may (1) start receiving care only when their disease is more severe, (2) experience faster disease progression even while receiving care, or (3) receive follow-up care less frequently conditional on disease severity. We show theoretically and empirically that failing to account for any of these disparities can result in biased estimates of severity (e.g., underestimating severity for disadvantaged groups). On a dataset of heart failure patients, we show that our model can identify groups that face each type of health disparity, and that accounting for these disparities while inferring disease severity meaningfully shifts which patients are considered high-risk.

Akshay Valsaraj, Ithihas Madala, Nikhil Garg et al.

Alzheimers disease is a fatal progressive brain disorder that worsens with time. It is high time we have inexpensive and quick clinical diagnostic techniques for early detection and care. In previous studies, various Machine Learning techniques and Pre-trained Deep Learning models have been used in conjunction with the extraction of various acoustic and linguistic features. Our study focuses on three models for the classification task in the ADReSS (The Alzheimers Dementia Recognition through Spontaneous Speech) 2021 Challenge. We use the well-balanced dataset provided by the ADReSS Challenge for training and validating our models. Model 1 uses various acoustic features from the eGeMAPs feature-set, Model 2 uses various linguistic features that we generated from auto-generated transcripts and Model 3 uses the auto-generated transcripts directly to extract features using a Pre-trained BERT and TF-IDF. These models are described in detail in the models section.

Lizy Kanungo, Nikhil Garg, Anish Bhobe et al.

This work proposes a hybrid Brain Computer Interface system for the automation of a wheelchair for the disabled. Herein a working prototype of a BCI-based wheelchair is detailed that can navigate inside a typical home environment with minimum structural modification and without any visual obstruction and discomfort to the user. The prototype is based on a combined mechanism of steady-state visually evoked potential and eye blinks. To elicit SSVEP, LEDs flickering at 13Hz and 15Hz were used to select the left and right direction, respectively, and EEG data was recorded. In addition, the occurrence of three continuous blinks was used as an indicator for stopping an ongoing action. The wavelet packet denoising method was applied, followed by feature extraction methods such as Wavelet Packet Decomposition and Canonical Correlation Analysis over narrowband reconstructed EEG signals. Bayesian optimization was used to obtain 5 fold cross-validations to optimize the hyperparameters of the Support Vector Machine. The resulting new model was tested and the average cross-validation accuracy 89.65% + 6.6% (SD) and testing accuracy 83.53% + 8.59% (SD) were obtained. The wheelchair was controlled by RaspberryPi through WiFi. The developed prototype demonstrated an average of 86.97% success rate for all trials with 4.015s for each command execution. The prototype can be used efficiently in a home environment without causing any discomfort to the user.

Nikhil Garg, Lodewijk Gelauff, Sukolsak Sakshuwong et al.

Elections and opinion polls often have many candidates, with the aim to either rank the candidates or identify a small set of winners according to voters' preferences. In practice, voters do not provide a full ranking; instead, each voter provides their favorite K candidates, potentially in ranked order. The election organizer must choose K and an aggregation rule. We provide a theoretical framework to make these choices. Each K-Approval or K-partial ranking mechanism (with a corresponding positional scoring rule) induces a learning rate for the speed at which the election correctly recovers the asymptotic outcome. Given the voter choice distribution, the election planner can thus identify the rate optimal mechanism. Earlier work in this area provides coarse order-of-magnitude guaranties which are not sufficient to make such choices. Our framework further resolves questions of when randomizing between multiple mechanisms may improve learning, for arbitrary voter noise models. Finally, we use data from 5 large participatory budgeting elections that we organized across several US cities, along with other ranking data, to demonstrate the utility of our methods. In particular, we find that historically such elections have set K too low and that picking the right mechanism can be the difference between identifying the ultimate winner with only a 80% probability or a 99.9% probability after 400 voters.

Nikhil Garg, James Henderson

We propose a Bayesian model of unsupervised semantic role induction in multiple languages, and use it to explore the usefulness of parallel corpora for this task. Our joint Bayesian model consists of individual models for each language plus additional latent variables that capture alignments between roles across languages. Because it is a generative Bayesian model, we can do evaluations in a variety of scenarios just by varying the inference procedure, without changing the model, thereby comparing the scenarios directly. We compare using only monolingual data, using a parallel corpus, using a parallel corpus with annotations in the other language, and using small amounts of annotation in the target language. We find that the biggest impact of adding a parallel corpus to training is actually the increase in mono-lingual data, with the alignments to another language resulting in small improvements, even with labeled data for the other language.