As language models evolve to tackle complex and multifaceted tasks, their evaluation must adapt to capture this intricacy.A granular, skillspecific understanding of model capabilities can empower researchers to make informed model development plans.In this paper, we introduce SKILLVERSE, an unsupervised treestructured diagnosis framework for understanding model proficiency in specific abilities.With LLM as a judge, SKILLVERSE first critiques the model responses, and then organizes them into a hierarchical structure termed dendrogram.Given proficiency at arbitrary levels of granularity, SKILLVERSE is flexible to produce insights of behaviors of modern large models.We also demonstrate its efficacy in two downstream tasks: 1) improving model in-context learning by 25% using a tree-search algorithm to select more informative few shots, and 2) accurately predicting new model weaknesses with a 55% success rate, 22% higher than the baseline.

Jiao Sun, Deqing Fu, Yushi Hu, Su Wang, Royi Rassin, Da-Cheng Juan, Dana Alon, Charles Herrmann, Sjoerd Van Steenkiste, Ranjay Krishna, Cyrus Rashtchian. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Recent Text-to-Image (T2I) generation models such as Stable Diffusion and Imagen have made significant progress in generating high-resolution images based on text descriptions. However, many generated images still suffer from issues such as artifacts/implausibility, misalignment with text descriptions, and low aesthetic quality. Inspired by the success of Reinforcement Learning with Human Feedback (RLHF) for large language models, prior works collected human-provided scores as feedback on generated images and trained a reward model to improve the T2I generation. In this paper, we enrich the feedback signal by (i) marking image regions that are implausible or misaligned with the text, and (ii) annotating which words in the text prompt are misrepresented or missing on the image. We collect such rich human feedback on 18K generated images (RichHF-18K) and train a multimodal transformer to predict the rich feedback automatically. We show that the predicted rich human feedback can be leveraged to improve image generation, for example, by selecting high-quality training data to finetune and improve the generative models, or by creating masks with predicted heatmaps to inpaint the problematic regions. Notably, the improvements generalize to models (Muse) beyond those used to generate the images on which human feedback data were collected (Stable Diffusion variants). The RichHF-18K data set will be released in our GitHub repository: https://github.com/google-research/google-research/tree/master/richhf_18k.

Jiao Sun, Thibault Sellam, Elizabeth Clark, Tu Vu, Timothy Dozat, Dan Garrette, Aditya Siddhant, Jacob Eisenstein, Sebastian Gehrmann. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Jiao Sun, Yufei Tian, Wangchunshu Zhou, Nan Xu, Qian Hu, Rahul Gupta, John Wieting, Nanyun Peng, Xuezhe Ma. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Large language models are trained in two stages: (1) unsupervised pretraining from raw text, to learn general-purpose representations, and (2) large scale instruction tuning and reinforcement learning, to better align to end tasks and user preferences. We measure the relative importance of these two stages by training LIMA, a 65B parameter LLaMa language model fine-tuned with the standard supervised loss on only 1,000 carefully curated prompts and responses, without any reinforcement learning or human preference modeling. LIMA demonstrates remarkably strong performance, learning to follow specific response formats from only a handful of examples in the training data, including complex queries that range from planning trip itineraries to speculating about alternate history. Moreover, the model tends to generalize well to unseen tasks that did not appear in the training data. In a controlled human study, responses from LIMA are either equivalent or strictly preferred to GPT-4 in 43\% of cases; this statistic is as high as 58\% when compared to Bard and 65\% versus DaVinci003, which was trained with human feedback. Taken together, these results strongly suggest that almost all knowledge in large language models is learned during pretraining, and only limited instruction tuning data is necessary to teach models to produce high quality output.

The goal of the crime forecasting problem is to predict different types of crimes for each geographical region (like a neighborhood or censor tract) in the near future. Since nearby regions usually have similar socioeconomic characteristics which indicate similar crime patterns, recent state-of-the-art solutions constructed a distance-based region graph and utilized Graph Neural Network (GNN) techniques for crime forecasting, because the GNN techniques could effectively exploit the latent relationships between neighboring region nodes in the graph if the edges reveal high dependency or correlation. However, this distance-based pre-defined graph can not fully capture crime correlation between regions that are far from each other but share similar crime patterns. Hence, to make a more accurate crime prediction, the main challenge is to learn a better graph that reveals the dependencies between regions in crime occurrences and meanwhile captures the temporal patterns from historical crime records. To address these challenges, we propose an end-to-end graph convolutional recurrent network called HAGEN with several novel designs for crime prediction. Specifically, our framework could jointly capture the crime correlation between regions and the temporal crime dynamics by combining an adaptive region graph learning module with the Diffusion Convolution Gated Recurrent Unit (DCGRU). Based on the homophily assumption of GNN (i.e., graph convolution works better where neighboring nodes share the same label), we propose a homophily-aware constraint to regularize the optimization of the region graph so that neighboring region nodes on the learned graph share similar crime patterns, thus fitting the mechanism of diffusion convolution. Empirical experiments and comprehensive analysis on two real-world datasets showcase the effectiveness of HAGEN.

People write personalized greeting cards on various occasions. While prior work has studied gender roles in greeting card messages, systematic analysis at scale and tools for raising the awareness of gender stereotyping remain under-investigated. To this end, we collect a large greeting card message corpus covering three different occasions (birthday, Valentine’s Day and wedding) from three sources (exemplars from greeting message websites, real-life greetings from social media and language model generated ones). We uncover a wide range of gender stereotypes in this corpus via topic modeling, odds ratio and Word Embedding Association Test (WEAT). We further conduct a survey to understand people’s perception of gender roles in messages from this corpus and if gender stereotyping is a concern. The results show that people want to be aware of gender roles in the messages, but remain unconcerned unless the perceived gender roles conflict with the recipient’s true personality. In response, we developed GreetA, an interactive visualization and writing assistant tool to visualize fine-grained topics in greeting card messages drafted by the users and the associated gender perception scores, but without suggesting text changes as an intervention.

Jiao Sun, Anjali Narayan-Chen, Shereen Oraby, Alessandra Cervone, Tagyoung Chung, Jing Huang, Yang Liu, Nanyun Peng. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Jiao Sun, Anjali Narayan-Chen, Shereen Oraby, Shuyang Gao, Tagyoung Chung, Jing Huang, Yang Liu, Nanyun Peng. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

What does it mean for a generative AI model to be explainable? The emergent discipline of explainable AI (XAI) has made great strides in helping people understand discriminative models. Less attention has been paid to generative models that produce artifacts, rather than decisions, as output. Meanwhile, generative AI (GenAI) technologies are maturing and being applied to application domains such as software engineering. Using scenario-based design and question-driven XAI design approaches, we explore users’ explainability needs for GenAI in three software engineering use cases: natural language to code, code translation, and code auto-completion. We conducted 9 workshops with 43 software engineers in which real examples from state-of-the-art generative AI models were used to elicit users’ explainability needs. Drawing from prior work, we also propose 4 types of XAI features for GenAI for code and gathered additional design ideas from participants. Our work explores explainability needs for GenAI for code and demonstrates how human-centered approaches can drive the technical development of XAI in novel domains.

Understanding how events are semantically related to each other is the essence of reading comprehension.Recent event-centric reading comprehension datasets focus mostly on event arguments or temporal relations.While these tasks partially evaluate machines' ability of narrative understanding, human-like reading comprehension requires the capability to process event-based information beyond arguments and temporal reasoning.For example, to understand causality between events, we need to infer motivation or purpose; to establish event hierarchy, we need to understand the composition of events.To facilitate these tasks, we introduce ESTER, a comprehensive machine reading comprehension (MRC) dataset for Event Semantic Relation Reasoning.The dataset leverages natural language queries to reason about the five most common event semantic relations, provides more than 6K questions, and captures 10.1K event relation pairs.Experimental results show that the current SOTA systems achieve 22.1%, 63.3% and 83.5% for token-based exact-match (EM), F 1 and event-based HIT@1 scores, which are all significantly below human performances (36.0%, 79.6%, 100% respectively), highlighting our dataset as a challenging benchmark.1

We propose to control paraphrase generation through carefully chosen target syntactic structures to generate more proper and higher quality paraphrases. Our model, AESOP, leverages a pretrained language model and adds deliberately chosen syntactical control via a retrieval-based selection module to generate fluent paraphrases. Experiments show that AESOP achieves state-of-the-art performances on semantic preservation and syntactic conformation on two benchmark datasets with ground-truth syntactic control from human-annotated exemplars. Moreover, with the retrieval-based target syntax selection module, AESOP generates paraphrases with even better qualities than the current best model using human-annotated target syntactic parses according to human evaluation. We further demonstrate the effectiveness of AESOP to improve classification models' robustness to syntactic perturbation by data augmentation on two GLUE tasks.

Deregulation of gene expression is associated with the pathogenesis of numerous human diseases including cancer. Current data analyses on gene expression are mostly focused on differential gene/transcript expression in big data-driven studies. However, a poor connection to the proteome changes is a widespread problem in current data analyses. This is partly due to the complexity of gene regulatory pathways at the post-transcriptional level. In this study, we overcome these limitations and introduce a graph-based learning model, PTNet, which simulates the microRNAs (miRNAs) that regulate gene expression post-transcriptionally in silico. Our model does not require large-scale proteomics studies to measure the protein expression and can successfully predict the protein levels by considering the miRNA-mRNA interaction network, the mRNA expression, and the miRNA expression. Large-scale experiments on simulations and real cancer high-throughput datasets using PTNet validated that (i) the miRNA-mediated interaction network affects the abundance of corresponding proteins and (ii) the predicted protein expression has a higher correlation with the proteomics data (ground-truth) than the mRNA expression data. The classification performance also shows that the predicted protein expression has an improved prediction power on cancer outcomes compared to the prediction done by the mRNA expression data only or considering both mRNA and miRNA. Availability: PTNet toolbox is available at http://github.com/CompbioLabUCF/PTNet.

MOTIVATION: Accurate disease phenotype prediction plays an important role in the treatment of heterogeneous diseases like cancer in the era of precision medicine. With the advent of high throughput technologies, more comprehensive multi-omics data is now available that can effectively link the genotype to phenotype. However, the interactive relation of multi-omics datasets makes it particularly challenging to incorporate different biological layers to discover the coherent biological signatures and predict phenotypic outcomes. In this study, we introduce omicsGAN, a generative adversarial network model to integrate two omics data and their interaction network. The model captures information from the interaction network as well as the two omics datasets and fuse them to generate synthetic data with better predictive signals. RESULTS: Large-scale experiments on The Cancer Genome Atlas breast cancer, lung cancer and ovarian cancer datasets validate that (i) the model can effectively integrate two omics data (e.g. mRNA and microRNA expression data) and their interaction network (e.g. microRNA-mRNA interaction network). The synthetic omics data generated by the proposed model has a better performance on cancer outcome classification and patients survival prediction compared to original omics datasets. (ii) The integrity of the interaction network plays a vital role in the generation of synthetic data with higher predictive quality. Using a random interaction network does not allow the framework to learn meaningful information from the omics datasets; therefore, results in synthetic data with weaker predictive signals. AVAILABILITY AND IMPLEMENTATION: Source code is available at: https://github.com/CompbioLabUCF/omicsGAN. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Online fraud is the well-known dark side of the modern Internet. Unsupervised fraud detection algorithms are widely used to address this problem. However, selecting features, adjusting hyperparameters, evaluating the algorithms, and eliminating false positives all require human expert involvement. In this work, we design and implement an end-to-end interactive visualization system, FDHelper, based on the deep understanding of the mechanism of the black market and fraud detection algorithms. We identify a workflow based on experience from both fraud detection algorithm experts and domain experts. Using a multi-granularity three-layer visualization map embedding an entropy-based distance metric ColDis, analysts can interactively select different feature sets, refine fraud detection algorithms, tune parameters and evaluate the detection result in near real-time. We demonstrate the effectiveness and significance of FDHelper through two case studies with state-of-the-art fraud detection algorithms, interviews with domain experts and algorithm experts, and a user study with eight first-time end users.

In this paper, we propose a high-throughput and energy-efficient Processing-in-Memory accelerator (PIM-Aligner) to execute DNA short read alignment based on an optimized and hardware-friendly alignment algorithm. We first reconstruct the existing sequence alignment algorithm based on BWT and FM-index such that it can be fully implemented in PIM platforms. It supports exact alignment and also handles mismatches to reduce excessive backtracking. We then develop PIM-Aligner platform that transforms SOT-MRAM array to a potential computational memory to accelerate the reconstructed alignment-in-memory algorithm incurring a low cost on top of original SOT-MRAM chips (less than 10% of chip area). Accordingly, we present a local data partitioning, mapping, and pipeline technique to maximize the parallelism in multiple computational sub-array while doing the alignment task. The simulation results show that PIM-Aligner outperforms recent platforms based on dynamic programming with ~ 3.1× higher throughput per Watt. Besides, PIM-Aligner improves the short read alignment throughput per Watt per mm2by ~ 9× and 1.9× compared to FM-index-based ASIC and processing-in-ReRAM designs, respectively.

MOTIVATION: Accurate estimation of transcript isoform abundance is critical for downstream transcriptome analyses and can lead to precise molecular mechanisms for understanding complex human diseases, like cancer. Simplex mRNA Sequencing (RNA-Seq) based isoform quantification approaches are facing the challenges of inherent sampling bias and unidentifiable read origins. A large-scale experiment shows that the consistency between RNA-Seq and other mRNA quantification platforms is relatively low at the isoform level compared to the gene level. In this project, we developed a platform-integrated model for transcript quantification (IntMTQ) to improve the performance of RNA-Seq on isoform expression estimation. IntMTQ, which benefits from the mRNA expressions reported by the other platforms, provides more precise RNA-Seq-based isoform quantification and leads to more accurate molecular signatures for disease phenotype prediction. RESULTS: In the experiments to assess the quality of isoform expression estimated by IntMTQ, we designed three tasks for clustering and classification of 46 cancer cell lines with four different mRNA quantification platforms, including newly developed NanoString's nCounter technology. The results demonstrate that the isoform expressions learned by IntMTQ consistently provide more and better molecular features for downstream analyses compared with five baseline algorithms which consider RNA-Seq data only. An independent RT-qPCR experiment on seven genes in twelve cancer cell lines showed that the IntMTQ improved overall transcript quantification. The platform-integrated algorithms could be applied to large-scale cancer studies, such as The Cancer Genome Atlas (TCGA), with both RNA-Seq and array-based platforms available. AVAILABILITY AND IMPLEMENTATION: Source code is available at: https://github.com/CompbioLabUcf/IntMTQ. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Classified as a complex big data analytics problem, DNA short read alignment serves as a major sequential bottleneck to massive amounts of data generated by next-generation sequencing platforms. With Von-Neumann computing architectures struggling to address such computationally-expensive and memory-intensive task today, Processing-in-Memory (PIM) platforms are gaining growing interests. In this paper, an energy-efficient and parallel PIM accelerator (AlignS) is proposed to execute DNA short read alignment based on an optimized and hardware-friendly alignment algorithm. We first develop AlignS platform that harnesses SOT-MRAM as computational memory and transforms it to a fundamental processing unit for short read alignment. Accordingly, we present a novel, customized, highly parallel read alignment algorithm that only seeks the proposed simple and parallel in-memory operations (i.e. comparisons and additions). AlignS is then optimized through a new correlated data partitioning and mapping methodology that allows local storage and processing of DNA sequence to fully exploit the algorithm-level's parallelism, and to accelerate both exact and inexact matches. The device-to-architecture co-simulation results show that AlignS improves the short read alignment throughput per Watt per mm2 by ~12× compared to the ASIC accelerator. Compared to recent FM-index-based ReRAM platform, AlignS achieves 1.6× higher throughput per Watt.

In this work, we present GraphS architecture, which transforms current Spin Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to massively parallel computational units capable of accelerating graph processing applications. GraphS can be leveraged to greatly reduce energy consumption dealing with underlying adjacency matrix computations, eliminating unnecessary off-chip accesses and providing ultra-high internal bandwidth. The device-to-architecture co-simulation for three social network data-sets indicate roughly 3.6× higher energy-efficiency and 5.3× speed-up over recent ReRAM crossbar. It achieves ~4× higher energy-efficiency and 5.1× speed-up over recent processing-in-DRAM acceleration methods.

Discovering fraud user behaviors is vital to keeping online websites healthy. Fraudsters usually exhibit grouping behaviors, and researchers have effectively leveraged this behavior to design unsupervised algorithms to detect fraud user groups. In this work, we propose a visualization system, FraudVis, to visually analyze the unsupervised fraud detection algorithms from temporal, intra-group correlation, inter-group correlation, feature selection, and the individual user perspectives. FraudVis helps domain experts better understand the algorithm output and the detected fraud behaviors. Meanwhile, FraudVis also helps algorithm experts to fine-tune the algorithm design through the visual comparison. By using the visualization system, we solve two real-world cases of fraud detection, one for a social video website and another for an e-commerce website. The results on both cases demonstrate the effectiveness of FraudVis in understanding unsupervised fraud detection algorithms.