Arun Iyer

Chanakya Ekbote, Ajinkya Pankaj Deshpande, Arun Iyer et al.

Unsupervised node representations learnt using contrastive learning-based methods have shown good performance on downstream tasks. However, these methods rely on augmentations that mimic low-pass filters, limiting their performance on tasks requiring different eigen-spectrum parts. This paper presents a simple filter-based augmentation method to capture different parts of the eigen-spectrum. We show significant improvements using these augmentations. Further, we show that sharing the same weights across these different filter augmentations is possible, reducing the computational load. In addition, previous works have shown that good performance on downstream tasks requires high dimensional representations. Working with high dimensions increases the computations, especially when multiple augmentations are involved. We mitigate this problem and recover good performance through lower dimensional embeddings using simple random Fourier feature projections. Our method, FiGURe achieves an average gain of up to 4.4%, compared to the state-of-the-art unsupervised models, across all datasets in consideration, both homophilic and heterophilic. Our code can be found at: https://github.com/microsoft/figure.

Pratik Ramesh, George Stoica, Arun Iyer et al.

Model merging has shown that multitask models can be created by directly combining the parameters of different models that are each specialized on tasks of interest. However, models trained independently on distinct tasks often exhibit interference that degrades the merged model's performance. To solve this problem, we formally define the notion of Cross-Task Interference as the drift in the representation of the merged model relative to its constituent models. Reducing cross-task interference is key to improving merging performance. To address this issue, we propose our method, Resolving Interference (RI), a light-weight adaptation framework which disentangles expert models to be functionally orthogonal to the space of other tasks, thereby reducing cross-task interference. RI does this whilst using only unlabeled auxiliary data as input (i.e., no task-data is needed), allowing it to be applied in data-scarce scenarios. RI consistently improves the performance of state-of-the-art merging methods by up to 3.8% and generalization to unseen domains by up to 2.3%. We also find RI to be robust to the source of auxiliary input while being significantly less sensitive to tuning of merging hyperparameters. Our codebase is available at: https://github.com/pramesh39/resolving_interference

Naman Gupta, Vaibhav Singh, Arun Iyer et al.

Sequential multi-agent reasoning frameworks such as Chain-of-Agents (CoA) handle long-context queries by decomposing inputs into chunks and processing them sequentially using LLM-based worker agents that read from and update a bounded shared memory. From a probabilistic perspective, CoA aims to approximate the conditional distribution corresponding to a model capable of jointly reasoning over the entire long context. CoA achieves this through a latent-state factorization in which only bounded summaries of previously processed evidence are passed between agents. The resulting bounded-memory approximation introduces a lossy information bottleneck, making the final evidence state inherently dependent on the order in which chunks are processed. In this work, we study the problem of chunk ordering for long-context reasoning. We use the well-known Chow-Liu trees to learn a dependency structure that prioritizes strongly related chunks. Empirically, we show that a breadth-first traversal of the resulting tree yields chunk orderings that reduce information loss across agents and consistently outperform both default document-chunk ordering and semantic score-based ordering in answer relevance and exact-match accuracy across three long-context benchmarks.

Ajinkya Deshpande, Anmol Agarwal, Shashank Shet et al.

LLMs have demonstrated significant potential in code generation tasks, achieving promising results at the function or statement level across various benchmarks. However, the complexities associated with creating code artifacts like classes, particularly within the context of real-world software repositories, remain underexplored. Prior research treats class-level generation as an isolated task, neglecting the intricate dependencies & interactions that characterize real-world software environments. To address this gap, we introduce RepoClassBench, a comprehensive benchmark designed to rigorously evaluate LLMs in generating complex, class-level code within real-world repositories. RepoClassBench includes "Natural Language to Class generation" tasks across Java, Python & C# from a selection of repositories. We ensure that each class in our dataset not only has cross-file dependencies within the repository but also includes corresponding test cases to verify its functionality. We find that current models struggle with the realistic challenges posed by our benchmark, primarily due to their limited exposure to relevant repository contexts. To address this shortcoming, we introduce Retrieve-Repotools-Reflect (RRR), a novel approach that equips LLMs with static analysis tools to iteratively navigate & reason about repository-level context in an agent-based framework. Our experiments demonstrate that RRR significantly outperforms existing baselines on RepoClassBench, showcasing its effectiveness across programming languages & under various settings. Our findings emphasize the critical need for code-generation benchmarks to incorporate repo-level dependencies to more accurately reflect the complexities of software development. Our work shows the benefits of leveraging specialized tools to enhance LLMs' understanding of repository context. We plan to make our dataset & evaluation harness public.

Shashank Kirtania, Naman Gupta, Priyanshu Gupta et al.

Large Language Models (LLMs) often generate incorrect or outdated information, especially in low-resource settings or when dealing with private data. To address this, Retrieval-Augmented Generation (RAG) uses external knowledge bases (KBs), but these can also suffer from inaccuracies. We introduce STACKFEED, a novel Structured Textual Actor-Critic Knowledge base editing with FEEDback approach that iteratively refines the KB based on expert feedback using a multi-actor, centralized critic reinforcement learning framework. STACKFEED defines a ReACT actor agent on each document to perform structured edits based on document specific targeted instructions. Experimental results showcase that STACKFEED significantly improves KB quality and performance of the RAG system. We evaluate STACKFEED on low-resource programming problems, modified python packaged and factual question-answering tasks.

Shashank Kirtania, Arun Iyer

Recent advances in automated theorem proving use Large Language Models (LLMs) to translate informal mathematical statements into formal proofs. However, informal cues are often ambiguous or lack strict logical structure, making it hard for models to interpret them precisely. While existing methods achieve strong performance, little is known about how LLMs internally represent informal cues, or how these influence proof generation. To address this, we explore \textit{activation steering}, an inference-time intervention that identifies linear directions in residual activations associated with informal reasoning traces and adjusts them to improve proof construction without fine-tuning. This mechanism also yields interpretable information about how reasoning is internally encoded in the activation space of LLMs. We test our method for generating formal proofs from already-formalized theorems. Our contributions are twofold: (1) a novel activation-based intervention for guiding proof synthesis in LLMs; and (2) demonstration that this intervention improves performance under two decoding strategies (sampling and best-first search) without any further training.

Naman Gupta, Shreeyash Gowaikar, Arun Iyer et al.

Reasoning over very long inputs remains difficult for large language models (LLMs). Common workarounds either shrink the input via retrieval (risking missed evidence), enlarge the context window (straining selectivity), or stage multiple agents to read in pieces. In staged pipelines (e.g., Chain of Agents, CoA), free-form summaries passed between agents can discard crucial details and amplify early mistakes. We introduce COSMIR (Chain Orchestrated Structured Memory for Iterative Reasoning), a chain-style framework that replaces ad hoc messages with a structured memory. A Planner agent first turns a user query into concrete, checkable sub-questions. worker agents process chunks via a fixed micro-cycle: Extract, Infer, Refine, writing all updates to the shared memory. A Manager agent then Synthesizes the final answer directly from the memory. This preserves step-wise read-then-reason benefits while changing both the communication medium (structured memory) and the worker procedure (fixed micro-cycle), yielding higher faithfulness, better long-range aggregation, and auditability. On long-context QA from the HELMET suite, COSMIR reduces propagation-stage information loss and improves accuracy over a CoA baseline.

Vijay Lingam, Chanakya Ekbote, Manan Sharma et al.

Graph Neural Networks (GNNs) exploit signals from node features and the input graph topology to improve node classification task performance. However, these models tend to perform poorly on heterophilic graphs, where connected nodes have different labels. Recently proposed GNNs work across graphs having varying levels of homophily. Among these, models relying on polynomial graph filters have shown promise. We observe that solutions to these polynomial graph filter models are also solutions to an overdetermined system of equations. It suggests that in some instances, the model needs to learn a reasonably high order polynomial. On investigation, we find the proposed models ineffective at learning such polynomials due to their designs. To mitigate this issue, we perform an eigendecomposition of the graph and propose to learn multiple adaptive polynomial filters acting on different subsets of the spectrum. We theoretically and empirically show that our proposed model learns a better filter, thereby improving classification accuracy. We study various aspects of our proposed model including, dependency on the number of eigencomponents utilized, latent polynomial filters learned, and performance of the individual polynomials on the node classification task. We further show that our model is scalable by evaluating over large graphs. Our model achieves performance gains of up to 5% over the state-of-the-art models and outperforms existing polynomial filter-based approaches in general.

Naman Jain, Skanda Vaidyanath, Arun Iyer et al.

Large pre-trained language models such as GPT-3, Codex, and Google's language model are now capable of generating code from natural language specifications of programmer intent. We view these developments with a mixture of optimism and caution. On the optimistic side, such large language models have the potential to improve productivity by providing an automated AI pair programmer for every programmer in the world. On the cautionary side, since these large language models do not understand program semantics, they offer no guarantees about quality of the suggested code. In this paper, we present an approach to augment these large language models with post-processing steps based on program analysis and synthesis techniques, that understand the syntax and semantics of programs. Further, we show that such techniques can make use of user feedback and improve with usage. We present our experiences from building and evaluating such a tool jigsaw, targeted at synthesizing code for using Python Pandas API using multi-modal inputs. Our experience suggests that as these large language models evolve for synthesizing code from intent, jigsaw has an important role to play in improving the accuracy of the systems.

Vijay Lingam, Rahul Ragesh, Arun Iyer et al.

Graph Neural Networks (GNNs) exhibit excellent performance when graphs have strong homophily property, i.e. connected nodes have the same labels. However, they perform poorly on heterophilic graphs. Several approaches address the issue of heterophily by proposing models that adapt the graph by optimizing task-specific loss function using labelled data. These adaptations are made either via attention or by attenuating or enhancing various low-frequency/high-frequency signals, as needed for the task at hand. More recent approaches adapt the eigenvalues of the graph. One important interpretation of this adaptation is that these models select/weigh the eigenvectors of the graph. Based on this interpretation, we present an eigendecomposition based approach and propose EigenNetwork models that improve the performance of GNNs on heterophilic graphs. Performance improvement is achieved by learning flexible graph adaptation functions that modulate the eigenvalues of the graph. Regularization of these functions via parameter sharing helps to improve the performance even more. Our approach achieves up to 11% improvement in performance over the state-of-the-art methods on heterophilic graphs.

Vijay Lingam, Rahul Ragesh, Arun Iyer et al.

Graph Neural Networks (GNNs) have shown excellent performance on graphs that exhibit strong homophily with respect to the node labels i.e. connected nodes have same labels. However, they perform poorly on heterophilic graphs. Recent approaches have typically modified aggregation schemes, designed adaptive graph filters, etc. to address this limitation. In spite of this, the performance on heterophilic graphs can still be poor. We propose a simple alternative method that exploits Truncated Singular Value Decomposition (TSVD) of topological structure and node features. Our approach achieves up to ~30% improvement in performance over state-of-the-art methods on heterophilic graphs. This work is an early investigation into methods that differ from aggregation based approaches. Our experimental results suggest that it might be important to explore other alternatives to aggregation methods for heterophilic setting.

Vijay Lingam, Arun Iyer, Rahul Ragesh

Graph Neural Networks have shown excellent performance on semi-supervised classification tasks. However, they assume access to a graph that may not be often available in practice. In the absence of any graph, constructing k-Nearest Neighbor (kNN) graphs from the given data have shown to give improvements when used with GNNs over other semi-supervised methods. This paper proposes a semi-supervised graph learning method for cases when there are no graphs available. This method learns a graph as a convex combination of the unsupervised kNN graph and a supervised label-affinity graph. The label-affinity graph directly captures all the nodes' label-affinity with the labeled nodes, i.e., how likely a node has the same label as the labeled nodes. This affinity measure contrasts with the kNN graph where the metric measures closeness in the feature space. Our experiments suggest that this approach gives close to or better performance (up to 1.5%), while being simpler and faster (up to 70x) to train, than state-of-the-art graph learning methods. We also conduct several experiments to highlight the importance of individual components and contrast them with state-of-the-art methods.

Rahul Ragesh, Sundararajan Sellamanickam, Vijay Lingam et al.

We consider the problem of learning latent features (aka embedding) for users and items in a recommendation setting. Given only a user-item interaction graph, the goal is to recommend items for each user. Traditional approaches employ matrix factorization-based collaborative filtering methods. Recent methods using graph convolutional networks (e.g., LightGCN) achieve state-of-the-art performance. They learn both user and item embedding. One major drawback of most existing methods is that they are not inductive; they do not generalize for users and items unseen during training. Besides, existing network models are quite complex, difficult to train and scale. Motivated by LightGCN, we propose a graph convolutional network modeling approach for collaborative filtering CF-GCN. We solely learn user embedding and derive item embedding using light variant CF-LGCN-U performing neighborhood aggregation, making it scalable due to reduced model complexity. CF-LGCN-U models naturally possess the inductive capability for new items, and we propose a simple solution to generalize for new users. We show how the proposed models are related to LightGCN. As a by-product, we suggest a simple solution to make LightGCN inductive. We perform comprehensive experiments on several benchmark datasets and demonstrate the capabilities of the proposed approach. Experimental results show that similar or better generalization performance is achievable than the state of the art methods in both transductive and inductive settings.

Rahul Ragesh, Sundararajan Sellamanickam, Arun Iyer et al.

We consider the problem of learning efficient and inductive graph convolutional networks for text classification with a large number of examples and features. Existing state-of-the-art graph embedding based methods such as predictive text embedding (PTE) and TextGCN have shortcomings in terms of predictive performance, scalability and inductive capability. To address these limitations, we propose a heterogeneous graph convolutional network (HeteGCN) modeling approach that unites the best aspects of PTE and TextGCN together. The main idea is to learn feature embeddings and derive document embeddings using a HeteGCN architecture with different graphs used across layers. We simplify TextGCN by dissecting into several HeteGCN models which (a) helps to study the usefulness of individual models and (b) offers flexibility in fusing learned embeddings from different models. In effect, the number of model parameters is reduced significantly, enabling faster training and improving performance in small labeled training set scenario. Our detailed experimental studies demonstrate the efficacy of the proposed approach.

Rahul Ragesh, Sundararajan Sellamanickam, Vijay Lingam et al.

Graph convolutional networks (GCNs) have gained popularity due to high performance achievable on several downstream tasks including node classification. Several architectural variants of these networks have been proposed and investigated with experimental studies in the literature. Motivated by a recent work on simplifying GCNs, we study the problem of designing other variants and propose a framework to compose networks using building blocks of GCN. The framework offers flexibility to compose and evaluate different networks using feature and/or label propagation networks, linear or non-linear networks, with each composition having different computational complexity. We conduct a detailed experimental study on several benchmark datasets with many variants and present observations from our evaluation. Our empirical experimental results suggest that several newly composed variants are useful alternatives to consider because they are as competitive as, or better than the original GCN.