Zaid Khan

Zaid Khan, Vijay Kumar BG, Samuel Schulter et al.

Finetuning a large vision language model (VLM) on a target dataset after large scale pretraining is a dominant paradigm in visual question answering (VQA). Datasets for specialized tasks such as knowledge-based VQA or VQA in non natural-image domains are orders of magnitude smaller than those for general-purpose VQA. While collecting additional labels for specialized tasks or domains can be challenging, unlabeled images are often available. We introduce SelTDA (Self-Taught Data Augmentation), a strategy for finetuning large VLMs on small-scale VQA datasets. SelTDA uses the VLM and target dataset to build a teacher model that can generate question-answer pseudolabels directly conditioned on an image alone, allowing us to pseudolabel unlabeled images. SelTDA then finetunes the initial VLM on the original dataset augmented with freshly pseudolabeled images. We describe a series of experiments showing that our self-taught data augmentation increases robustness to adversarially searched questions, counterfactual examples and rephrasings, improves domain generalization, and results in greater retention of numerical reasoning skills. The proposed strategy requires no additional annotations or architectural modifications, and is compatible with any modern encoder-decoder multimodal transformer. Code available at https://github.com/codezakh/SelTDA.

Zaid Khan, Yun Fu

Contrastive vision-language models (e.g. CLIP) are typically created by updating all the parameters of a vision model and language model through contrastive training. Can such models be created by a small number of parameter updates to an already-trained language model and vision model? The literature describes techniques that can create vision-language models by updating a small number of parameters in a language model, but these require already aligned visual representations and are non-contrastive, hence unusable for latency-sensitive applications such as neural search. We explore the feasibility and benefits of parameter-efficient contrastive vision-language alignment through transfer learning: creating a model such as CLIP by minimally updating an already-trained vision and language model. We find that a minimal set of parameter updates ($<$7%) can achieve the same performance as full-model training, and updating specific components ($<$1% of parameters) can match 75% of full-model training. We describe a series of experiments: we show that existing knowledge is conserved more strongly in parameter-efficient training and that parameter-efficient scaling scales with model and dataset size. Where paired-image text data is scarce but strong multilingual language models exist (e.g. low resource languages), parameter-efficient training is even preferable to full-model training. Given a fixed compute budget, parameter-efficient training allows training larger models on the same hardware, achieving equivalent performance in less time. Parameter-efficient training hence constitutes an energy-efficient and effective training strategy for contrastive vision-language models that may be preferable to the full-model training paradigm for common use cases. Code and weights at https://github.com/codezakh/LilT.

Zaid Khan, Vijay Kumar BG, Xiang Yu et al.

Self-supervised vision-language pretraining from pure images and text with a contrastive loss is effective, but ignores fine-grained alignment due to a dual-stream architecture that aligns image and text representations only on a global level. Earlier, supervised, non-contrastive methods were capable of finer-grained alignment, but required dense annotations that were not scalable. We propose a single stream architecture that aligns images and language at multiple levels: global, fine-grained patch-token, and conceptual/semantic, using two novel tasks: symmetric cross-modality reconstruction (XMM) and a pseudo-labeled key word prediction (PSL). In XMM, we mask input tokens from one modality and use cross-modal information to reconstruct the masked token, thus improving fine-grained alignment between the two modalities. In PSL, we use attention to select keywords in a caption, use a momentum encoder to recommend other important keywords that are missing from the caption but represented in the image, and then train the visual encoder to predict the presence of those keywords, helping it learn semantic concepts that are essential for grounding a textual token to an image region. We demonstrate competitive performance and improved data efficiency on image-text retrieval, grounding, visual question answering/reasoning against larger models and models trained on more data. Code and models available at zaidkhan.me/SIMLA.

Zaid Khan, Justin Chih-Yao Chen, Jaemin Cho et al.

GPU kernels are the workhorse of modern deep learning, and optimizing them (via evolutionary search or coding agents) usually requires repeated measurement on target hardware. While these measurements provide the ground-truth signal necessary for kernel search, they are costly, because each evaluation of a kernel requires compilation and repeated execution on a GPU. As improvements in LLM inference reduce the cost of writing novel kernels and LLM-driven searches scale to large search budgets, on-device evaluation becomes a bottleneck. To address this, we study how LLMs can serve as selective GPU surrogates for kernel evaluation, by forecasting the performance of proposed kernels. A useful surrogate should be accurate, and it should be selective, by knowing when it could be wrong, and deferring to the GPU. To evaluate surrogates, we measure whether their forecasts are accurate, calibrated, and practically useful for recovering fast kernels under limited GPU-measurement budgets. Next, we study whether reinforcement learning can improve forecast accuracy and confidence calibration. Our experiments demonstrate that LLMs can accurately forecast relative kernel performance, that their utility can be improved through reinforcement learning. Used inside a kernel search, the surrogate lets the search consider several times as many candidates under the same GPU evaluation budget, and that leads to finding faster kernels than an equal-budget baseline. These results suggest that LLMs can play a broader role in kernel optimization, by acting as virtual models of a GPU rather than solely as kernel generators for search.

Zaid Khan, Vijay Kumar BG, Samuel Schulter et al.

Visual question answering (VQA) has traditionally been treated as a single-step task where each question receives the same amount of effort, unlike natural human question-answering strategies. We explore a question decomposition strategy for VQA to overcome this limitation. We probe the ability of recently developed large vision-language models to use human-written decompositions and produce their own decompositions of visual questions, finding they are capable of learning both tasks from demonstrations alone. However, we show that naive application of model-written decompositions can hurt performance. We introduce a model-driven selective decomposition approach for second-guessing predictions and correcting errors, and validate its effectiveness on eight VQA tasks across three domains, showing consistent improvements in accuracy, including improvements of >20% on medical VQA datasets and boosting the zero-shot performance of BLIP-2 above chance on a VQA reformulation of the challenging Winoground task. Project Site: https://zaidkhan.me/decomposition-0shot-vqa/

Hanqi Xiao, Vaidehi Patil, Zaid Khan et al.

As large language models (LLMs) become the engine behind conversational systems, their ability to reason about the intentions and states of their dialogue partners (i.e., form and use a theory-of-mind, or ToM) becomes increasingly critical for safe interaction with potentially adversarial partners. We propose a novel privacy-themed ToM challenge, ToM for Steering Beliefs (ToM-SB), in which a defender must act as a Double Agent to steer the beliefs of an attacker with partial prior knowledge within a shared universe. To succeed on ToM-SB, the defender must engage with and form a ToM of the attacker, with a goal of fooling the attacker into believing they have succeeded in extracting sensitive information. We find that strong frontier models like Gemini3-Pro and GPT-5.4 struggle on ToM-SB, often failing to fool attackers in hard scenarios with partial attacker prior knowledge, even when prompted to reason about the attacker's beliefs (ToM prompting). To close this gap, we train models on ToM-SB to act as AI Double Agents using reinforcement learning, testing both fooling and ToM rewards. Notably, we find a bidirectionally emergent relationship between ToM and attacker-fooling: rewarding fooling success alone improves ToM, and rewarding ToM alone improves fooling. Across four attackers with different strengths, six defender methods, and both in-distribution and out-of-distribution (OOD) evaluation, we find that gains in ToM and attacker-fooling are well-correlated, highlighting belief modeling as a key driver of success on ToM-SB. AI Double Agents that combine both ToM and fooling rewards yield the strongest fooling and ToM performance, outperforming Gemini3-Pro and GPT-5.4 with ToM prompting on hard scenarios. We also show that ToM-SB and AI Double Agents can be extended to stronger attackers, demonstrating generalization to OOD settings and the upgradability of our task.

Etash Guha, Ryan Marten, Sedrick Keh et al. · cmu

Reasoning models have made rapid progress on many benchmarks involving math, code, and science. Yet, there are still many open questions about the best training recipes for reasoning since state-of-the-art models often rely on proprietary datasets with little to no public information available. To address this, the goal of the OpenThoughts project is to create open-source datasets for training reasoning models. After initial explorations, our OpenThoughts2-1M dataset led to OpenThinker2-32B, the first model trained on public reasoning data to match DeepSeek-R1-Distill-32B on standard reasoning benchmarks such as AIME and LiveCodeBench. We then improve our dataset further by systematically investigating each step of our data generation pipeline with 1,000+ controlled experiments, which led to OpenThoughts3. Scaling the pipeline to 1.2M examples and using QwQ-32B as teacher yields our OpenThoughts3-7B model, which achieves state-of-the-art results: 53% on AIME 2025, 51% on LiveCodeBench 06/24-01/25, and 54% on GPQA Diamond - improvements of 15.3, 17.2, and 20.5 percentage points compared to the DeepSeek-R1-Distill-Qwen-7B. All of our datasets and models are available on https://openthoughts.ai.

Duy Nguyen, Hanqi Xiao, Archiki Prasad et al.

Self-distillation enables language models to learn on-policy from their own trajectories by using the same model as both student and teacher, with the teacher being conditioned on privileged information unavailable to the student. Such information can come in different types or views, such as solutions, demonstrations, feedback, or final answers. This setup provides dense token-level feedback without relying on a separate external model, but creates a fundamental asymmetry: the teacher may rely on view-specific information that the student cannot access at inference time. Moreover, the best type of privileged information is often task-dependent, making it difficult to choose a single teacher view. In this work, we address both these challenges jointly by introducing AVSD (Adaptive-View Self-Distillation), a novel method of self-distillation with multiple privileged-information views, which reconstructs token-level supervision by separating stable cross-view consensus from view-specific residual signals. AVSD identifies the consensus signal shared across views, which provides a reliable update direction, and then selectively adds the view-specific residual signal to adjust the update magnitude when it both aligns with the consensus direction and remains proportionate to the consensus signal. Experiments on math competition benchmarks (AIME24, AIME25, and HMMT25) show that AVSD consistently outperforms both single-view self-distillation baselines and GRPO, achieving average Avg@8 gains of 3.1% and 2.2% over the strongest baselines on Qwen3-8B and Qwen3-4B, respectively. Moreover, on code-generation benchmarks (Codeforces, LiveCodeBench v6) using Qwen3-8B, AVSD outperforms the single-view self-distillation baseline by 2.4% on average.

Hyunji Lee, Justin Chih-Yao Chen, Joykirat Singh et al.

Real-world agents operate over long and evolving horizons, where information is repeatedly updated and may interfere across memories, requiring accurate recall and aggregated reasoning over multiple pieces of information. However, existing benchmarks focus on static, independent recall and fail to capture these dynamic interactions between evolving memories. In this paper, we study how current memory-augmented agents perform in realistic, interference-heavy, long-horizon settings across diverse domains and question types. We introduce LongMINT (Long-Horizon Memory under INTerference), a benchmark featuring (1) long, highly interconnected contexts with frequently updated information that induces substantial interference, (2) diverse domains (state tracking, multi-turn dialogue, Wikipedia revisions, and GitHub commits), enabling evaluation of domain generalization, and (3) diverse question types that assess robustness to interference, including (i) single-target recall tasks requiring retrieval of a specific target from long contexts, and (ii) multi-target aggregation tasks requiring reasoning over multiple relevant pieces of information. Overall, LongMINT has 15.6k question-answering pairs over long-horizon contexts averaging 138.8k tokens and extending up to 1.8M tokens per instance. We evaluate 7 representative systems, including vanilla long-context LLMs, RAG, and memory-augmented agent frameworks. Across all systems, we observe consistently low performance (avg. 27.9% accuracy), especially on questions requiring aggregated reasoning over multiple pieces of evidence. Our analysis shows that performance is primarily limited by retrieval and memory construction. Furthermore, current memory systems struggle to recall and reason over earlier facts that are later revised or interfered with by subsequent context, with performance degrading as the number of intervening updates increases.

Joykirat Singh, Zaid Khan, Archiki Prasad et al.

Large language models (LLMs) are increasingly deployed on long-horizon tasks in partially observable environments, where they must act while inferring and tracking a complex environment state over many steps. This leads to two challenges: partial observability requires maintaining uncertainty over unobserved world attributes, and long interaction history causes context to grow without bound, diluting task-relevant information. A principled solution to both challenges is a belief state: a posterior distribution over environment states given past observations and actions, which compactly encodes history for decision making regardless of episode length. In LLM agents, however, the open-ended nature of text makes it unclear how to represent such a distribution. Therefore, we introduce Agent-BRACE: Agent Belief state Representation via Abstraction and Confidence Estimation, a method that decouples an LLM agent into a belief state model and a policy model, jointly optimized via reinforcement learning. The belief state model produces a structured approximation of the belief distribution: a set of atomic natural language claims about the environment, each annotated with an ordinal verbalized certainty label ranging from certain to unknown. The policy model conditions on this compact, structured approximate belief rather than the full history, learning to select actions under explicit uncertainty. Across long-horizon, partially observable embodied language environments, Agent-BRACE achieves an average absolute improvement of +14.5% (Qwen2.5-3B-Instruct) and +5.3% (Qwen3-4B-Instruct), outperforming strong RL baselines while maintaining a near-constant context window independent of episode length. Further analysis shows that the learned belief becomes increasingly calibrated over the course of an episode as evidence accumulates.

Jaewoo Lee, Archiki Prasad, Justin Chih-Yao Chen et al.

Information-seeking is a core capability for AI agents, requiring them to gather and reason over tool-generated information across long trajectories. However, such multi-step information-seeking tasks remain challenging for agents backed by language models. While process reward models (PRMs) can guide agents by ranking candidate steps at test-time, existing PRMs, designed for short reasoning with binary judgment, cannot capture richer dimensions of information-seeking steps, such as tool interactions and reasoning over tool outputs, nor handle the rapidly growing context in long-horizon tasks. To address these limitations, we introduce PRInTS, a generative PRM trained with dual capabilities: (1) dense scoring based on the PRM's reasoning across multiple step quality dimensions (e.g., interpretation of tool outputs, tool call informativeness) and (2) trajectory summarization that compresses the growing context while preserving essential information for step evaluation. Extensive evaluations across FRAMES, GAIA (levels 1-3), and WebWalkerQA (easy-hard) benchmarks on multiple models, along with ablations, reveal that best-of-n sampling with PRInTS enhances information-seeking abilities of open-source models as well as specialized agents, matching or surpassing the performance of frontier models with a much smaller backbone agent and outperforming other strong reward modeling baselines.

Zaid Khan, Yun Fu

Multimodal target/aspect sentiment classification combines multimodal sentiment analysis and aspect/target sentiment classification. The goal of the task is to combine vision and language to understand the sentiment towards a target entity in a sentence. Twitter is an ideal setting for the task because it is inherently multimodal, highly emotional, and affects real world events. However, multimodal tweets are short and accompanied by complex, possibly irrelevant images. We introduce a two-stream model that translates images in input space using an object-aware transformer followed by a single-pass non-autoregressive text generation approach. We then leverage the translation to construct an auxiliary sentence that provides multimodal information to a language model. Our approach increases the amount of text available to the language model and distills the object-level information in complex images. We achieve state-of-the-art performance on two multimodal Twitter datasets without modifying the internals of the language model to accept multimodal data, demonstrating the effectiveness of our translation. In addition, we explain a failure mode of a popular approach for aspect sentiment analysis when applied to tweets. Our code is available at \textcolor{blue}{\url{https://github.com/codezakh/exploiting-BERT-thru-translation}}.

Zaid Khan, Yun Fu

The goal of selective prediction is to allow an a model to abstain when it may not be able to deliver a reliable prediction, which is important in safety-critical contexts. Existing approaches to selective prediction typically require access to the internals of a model, require retraining a model or study only unimodal models. However, the most powerful models (e.g. GPT-4) are typically only available as black boxes with inaccessible internals, are not retrainable by end-users, and are frequently used for multimodal tasks. We study the possibility of selective prediction for vision-language models in a realistic, black-box setting. We propose using the principle of \textit{neighborhood consistency} to identify unreliable responses from a black-box vision-language model in question answering tasks. We hypothesize that given only a visual question and model response, the consistency of the model's responses over the neighborhood of a visual question will indicate reliability. It is impossible to directly sample neighbors in feature space in a black-box setting. Instead, we show that it is possible to use a smaller proxy model to approximately sample from the neighborhood. We find that neighborhood consistency can be used to identify model responses to visual questions that are likely unreliable, even in adversarial settings or settings that are out-of-distribution to the proxy model.

Zaid Khan, Vijay Kumar BG, Samuel Schulter et al.

Visual program synthesis is a promising approach to exploit the reasoning abilities of large language models for compositional computer vision tasks. Previous work has used few-shot prompting with frozen LLMs to synthesize visual programs. Training an LLM to write better visual programs is an attractive prospect, but it is unclear how to accomplish this. No dataset of visual programs for training exists, and acquisition of a visual program dataset cannot be easily crowdsourced due to the need for expert annotators. To get around the lack of direct supervision, we explore improving the program synthesis abilities of an LLM using feedback from interactive experience. We propose a method where we exploit existing annotations for a vision-language task to improvise a coarse reward signal for that task, treat the LLM as a policy, and apply reinforced self-training to improve the visual program synthesis ability of the LLM for that task. We describe a series of experiments on object detection, compositional visual question answering, and image-text retrieval, and show that in each case, the self-trained LLM outperforms or performs on par with few-shot frozen LLMs that are an order of magnitude larger. Website: https://zaidkhan.me/ViReP

Archiki Prasad, Elias Stengel-Eskin, Justin Chih-Yao Chen et al.

Unit tests (UTs) play an instrumental role in assessing code correctness as well as providing feedback to large language models (LLMs), motivating automated test generation. However, we uncover a trade-off between generating unit test inputs that reveal errors when given a faulty code and correctly predicting the unit test output without access to the gold solution. To address this trade-off, we propose UTGen, which teaches LLMs to generate unit test inputs that reveal errors along with their correct expected outputs based on task descriptions. Since model-generated tests can provide noisy signals (e.g., from incorrectly predicted outputs), we propose UTDebug that (i) scales UTGen via test-time compute to improve UT output prediction, and (ii) validates and backtracks edits based on multiple generated UTs to avoid overfitting, and helps LLMs debug effectively. We show that UTGen outperforms other LLM-based baselines by 7.59% based on a metric measuring the presence of both error-revealing UT inputs and correct UT outputs. When used with UTDebug, we find that feedback from UTGen's unit tests improves pass@1 accuracy of Qwen2.5 32B on HumanEvalFix and our own harder debugging split of MBPP+ by over 3.17% and 12.35% (respectively) over other LLM-based UT generation baselines. Moreover, we observe that feedback from Qwen2.5 32B-based UTGen model can enhance debugging with frontier LLMs like GPT-4o by 13.8%. Lastly, we demonstrate that UTGen is a better judge for code correctness, outperforming a state-of-the-art trained 8B reward model by 4.43% on HumanEval+ with best-of-10 sampling using Qwen2.5 7B.

Fucai Ke, Vijay Kumar B G, Xingjian Leng et al.

Visual reasoning (VR), which is crucial in many fields for enabling human-like visual understanding, remains highly challenging. Recently, compositional visual reasoning approaches, which leverage the reasoning abilities of large language models (LLMs) with integrated tools to solve problems, have shown promise as more effective strategies than end-to-end VR methods. However, these approaches face limitations, as frozen LLMs lack tool awareness in VR, leading to performance bottlenecks. While leveraging LLMs for reasoning is widely used in other domains, they are not directly applicable to VR due to limited training data, imperfect tools that introduce errors and reduce data collection efficiency in VR, and challenging in fine-tuning on noisy workflows. To address these challenges, we propose DWIM: i) Discrepancy-aware training Workflow generation, which assesses tool usage and extracts more viable workflows for training; and ii) Instruct-Masking fine-tuning, which guides the model to only clone effective actions, enabling the generation of more practical solutions. Our experiments demonstrate that DWIM achieves state-of-the-art performance across various VR tasks, exhibiting strong generalization on multiple widely-used datasets.

Justin Chih-Yao Chen, Archiki Prasad, Zaid Khan et al.

Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning abilities of LLMs, yet a fundamental limitation remains: models cannot learn from problems that are too difficult to solve under their current policy, as these yield no meaningful reward signal. We propose a simple yet effective solution based on task reformulation. We transform challenging open-ended problems into cognitively simpler variants -- such as multiple-choice and cloze formats -- that preserve the original answer while reducing the effective search space and providing denser learning signals. These reformulations span a spectrum from discriminative to generative tasks, which we exploit to bootstrap learning: models first learn from structured, easier formats, and this knowledge transfers back to improve performance on the original open-ended problems. Building on this insight, we introduce Cog-DRIFT, a framework that constructs reformulated variants and organizes them into an adaptive curriculum based on difficulty. Training progresses from easier to harder formats, enabling the model to learn from problems that previously yielded zero signal under standard RL post-training. Cog-DRIFT not only improves on the originally unsolvable hard problems (absolute +10.11% for Qwen and +8.64% for Llama) but also generalizes well to other held-out datasets. Across 2 models and 6 reasoning benchmarks, our method consistently outperforms standard GRPO and strong guided-exploration baselines. On average, Cog-DRIFT shows +4.72% (Qwen) and +3.23% (Llama) improvements over the second-best baseline. We further show that Cog-DRIFT improves pass@k at test time, and the curriculum improves sample efficiency. Overall, our results highlight task reformulation and curriculum learning as an effective paradigm for overcoming the exploration barrier in LLM post-training.

Zaid Khan, Archiki Prasad, Elias Stengel-Eskin et al. · allen-ai

Symbolic world modeling requires inferring and representing an environment's transitional dynamics as an executable program. Prior work has focused on largely deterministic environments with abundant interaction data, simple mechanics, and human guidance. We address a more realistic and challenging setting, learning in a complex, stochastic environment where the agent has only "one life" to explore a hostile environment without human guidance. We introduce OneLife, a framework that models world dynamics through conditionally-activated programmatic laws within a probabilistic programming framework. Each law operates through a precondition-effect structure, activating in relevant world states. This creates a dynamic computation graph that routes inference and optimization only through relevant laws, avoiding scaling challenges when all laws contribute to predictions about a complex, hierarchical state, and enabling the learning of stochastic dynamics even with sparse rule activation. To evaluate our approach under these demanding constraints, we introduce a new evaluation protocol that measures (a) state ranking, the ability to distinguish plausible future states from implausible ones, and (b) state fidelity, the ability to generate future states that closely resemble reality. We develop and evaluate our framework on Crafter-OO, our reimplementation of the Crafter environment that exposes a structured, object-oriented symbolic state and a pure transition function that operates on that state alone. OneLife can successfully learn key environment dynamics from minimal, unguided interaction, outperforming a strong baseline on 16 out of 23 scenarios tested. We also test OneLife's planning ability, with simulated rollouts successfully identifying superior strategies. Our work establishes a foundation for autonomously constructing programmatic world models of unknown, complex environments.

Zaid Khan, Elias Stengel-Eskin, Archiki Prasad et al. · allen-ai

Scientists often infer abstract procedures from specific instances of problems and use the abstractions to generate new, related instances. For example, programs encoding the formal rules and properties of a system have been useful in fields ranging from reinforcement learning (procedural environments) to physics (simulation engines). These programs can be seen as functions which execute to different outputs based on their parameterizations (e.g., gridworld configuration or initial physical conditions). We introduce the term EFA (Executable Functional Abstraction) to denote such programs for math problems. EFA-like constructs have been shown to be useful for mathematical reasoning as problem generators for stress-testing models. However, prior work has been limited to automatically constructing abstractions for grade-school math (whose simple rules are easy to encode in programs), while generating EFAs for advanced math has thus far required human engineering. We explore the automatic construction of EFAs for advanced mathematics problems by developing EFAGen, which operationalizes the task of automatically inferring an EFA for a given seed problem and solution as a program synthesis task. We first formalize the properties of any valid EFA as executable unit tests. Using execution feedback from the unit tests, we search over candidate programs sampled from a LLM to find EFA programs that are faithful to the generalized problem and solution class underlying the seed problem. We then apply the tests as a reward signal, training LLMs to become better writers of EFAs. We show that EFAs inferred by EFAGen are faithful to the seed problems, produce learnable problem variations, and that EFAGen can infer EFAs across diverse sources of competition-level math problems. Finally, we show uses of model-written EFAs e.g., finding harder/easier problem variants, as well as data generation.

Zaid Khan, Ali Farhadi, Ranjay Krishna et al. · allen-ai

When a human requests an LLM to complete a coding task using functionality from a large code repository, how do we provide context from the repo to the LLM? One approach is to add the entire repo to the LLM's context window. However, most tasks involve only fraction of symbols from a repo, longer contexts are detrimental to the LLM's reasoning abilities, and context windows are not unlimited. Alternatively, we could emulate the human ability to navigate a large repo, pick out the right functionality, and form a plan to solve the task. We propose MutaGReP (Mutation-guided Grounded Repository Plan Search), an approach to search for plans that decompose a user request into natural language steps grounded in the codebase. MutaGReP performs neural tree search in plan space, exploring by mutating plans and using a symbol retriever for grounding. On the challenging LongCodeArena benchmark, our plans use less than 5% of the 128K context window for GPT-4o but rival the coding performance of GPT-4o with a context window filled with the repo. Plans produced by MutaGReP allow Qwen 2.5 Coder 32B and 72B to match the performance of GPT-4o with full repo context and enable progress on the hardest LongCodeArena tasks. Project page: zaidkhan.me/MutaGReP

Zaid Khan, Yun Fu

Computer vision is widely deployed, has highly visible, society altering applications, and documented problems with bias and representation. Datasets are critical for benchmarking progress in fair computer vision, and often employ broad racial categories as population groups for measuring group fairness. Similarly, diversity is often measured in computer vision datasets by ascribing and counting categorical race labels. However, racial categories are ill-defined, unstable temporally and geographically, and have a problematic history of scientific use. Although the racial categories used across datasets are superficially similar, the complexity of human race perception suggests the racial system encoded by one dataset may be substantially inconsistent with another. Using the insight that a classifier can learn the racial system encoded by a dataset, we conduct an empirical study of computer vision datasets supplying categorical race labels for face images to determine the cross-dataset consistency and generalization of racial categories. We find that each dataset encodes a substantially unique racial system, despite nominally equivalent racial categories, and some racial categories are systemically less consistent than others across datasets. We find evidence that racial categories encode stereotypes, and exclude ethnic groups from categories on the basis of nonconformity to stereotypes. Representing a billion humans under one racial category may obscure disparities and create new ones by encoding stereotypes of racial systems. The difficulty of adequately converting the abstract concept of race into a tool for measuring fairness underscores the need for a method more flexible and culturally aware than racial categories.

Joseph P. Robinson, Zaid Khan, Yu Yin et al.

Kinship, a soft biometric detectable in media, is fundamental for a myriad of use-cases. Despite the difficulty of detecting kinship, annual data challenges using still-images have consistently improved performances and attracted new researchers. Now, systems reach performance levels unforeseeable a decade ago, closing in on performances acceptable to deploy in practice. Like other biometric tasks, we expect systems can receive help from other modalities. We hypothesize that adding modalities to FIW, which has only still-images, will improve performance. Thus, to narrow the gap between research and reality and enhance the power of kinship recognition systems, we extend FIW with multimedia (MM) data (i.e., video, audio, and text captions). Specifically, we introduce the first publicly available multi-task MM kinship dataset. To build FIW MM, we developed machinery to automatically collect, annotate, and prepare the data, requiring minimal human input and no financial cost. The proposed MM corpus allows the problem statements to be more realistic template-based protocols. We show significant improvements in all benchmarks with the added modalities. The results highlight edge cases to inspire future research with different areas of improvement. FIW MM supplies the data needed to increase the potential of automated systems to detect kinship in MM. It also allows experts from diverse fields to collaborate in novel ways.

Joseph P. Robinson, Yu Yin, Zaid Khan et al.

Recognizing Families In the Wild (RFIW): an annual large-scale, multi-track automatic kinship recognition evaluation that supports various visual kin-based problems on scales much higher than ever before. Organized in conjunction with the 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG) as a Challenge, RFIW provides a platform for publishing original work and the gathering of experts for a discussion of the next steps. This paper summarizes the supported tasks (i.e., kinship verification, tri-subject verification, and search & retrieval of missing children) in the evaluation protocols, which include the practical motivation, technical background, data splits, metrics, and benchmark results. Furthermore, top submissions (i.e., leader-board stats) are listed and reviewed as a high-level analysis on the state of the problem. In the end, the purpose of this paper is to describe the 2020 RFIW challenge, end-to-end, along with forecasts in promising future directions.