Deepak Nathani

Liangming Pan, Michael Saxon, Wenda Xu et al. · pku

Large language models (LLMs) have demonstrated remarkable performance across a wide array of NLP tasks. However, their efficacy is undermined by undesired and inconsistent behaviors, including hallucination, unfaithful reasoning, and toxic content. A promising approach to rectify these flaws is self-correction, where the LLM itself is prompted or guided to fix problems in its own output. Techniques leveraging automated feedback -- either produced by the LLM itself or some external system -- are of particular interest as they are a promising way to make LLM-based solutions more practical and deployable with minimal human feedback. This paper presents a comprehensive review of this emerging class of techniques. We analyze and taxonomize a wide array of recent work utilizing these strategies, including training-time, generation-time, and post-hoc correction. We also summarize the major applications of this strategy and conclude by discussing future directions and challenges.

Alexander D. Goldie, Zilin Wang, Adrian Hayler et al.

Automating the development of machine learning algorithms has the potential to unlock new breakthroughs. However, our ability to improve and evaluate algorithm discovery systems has thus far been limited by existing task suites. They suffer from many issues, such as: poor evaluation methodologies; data contamination; and containing saturated or very similar problems. Here, we introduce DiscoGen, a procedural generator of algorithm discovery tasks for machine learning, such as developing optimisers for reinforcement learning or loss functions for image classification. Motivated by the success of procedural generation in reinforcement learning, DiscoGen spans millions of tasks of varying difficulty and complexity from a range of machine learning fields. These tasks are specified by a small number of configuration parameters and can be used to optimise algorithm discovery agents (ADAs). We present DiscoBench, a benchmark consisting of a fixed, small subset of DiscoGen tasks for principled evaluation of ADAs. Finally, we propose a number of ambitious, impactful research directions enabled by DiscoGen, in addition to experiments demonstrating its use for prompt optimisation of an ADA. DiscoGen is released open-source at https://github.com/AlexGoldie/discogen.

Tengxiao Liu, Deepak Nathani, Zekun Li et al.

Recent progress in large language model (LLM) reasoning has focused on domains like mathematics and coding, where abundant high-quality data and objective evaluation metrics are readily available. In contrast, progress in LLM reasoning models remains limited in scientific domains such as medicine and materials science due to limited dataset coverage and the inherent complexity of open-ended scientific questions. To address these challenges, we introduce WildSci, a new dataset of domain-specific science questions automatically synthesized from peer-reviewed literature, covering 9 scientific disciplines and 26 subdomains. By framing complex scientific reasoning tasks in a multiple-choice format, we enable scalable training with well-defined reward signals. We further apply reinforcement learning to finetune models on these data and analyze the resulting training dynamics, including domain-specific performance changes, response behaviors, and generalization trends. Experiments on a suite of scientific benchmarks demonstrate the effectiveness of our dataset and approach. We release WildSci to enable scalable and sustainable research in scientific reasoning, available at https://huggingface.co/datasets/JustinTX/WildSci.

Deepak Nathani, David Wang, Liangming Pan et al.

Language Models (LMs) have shown impressive performance in various natural language tasks. However, when it comes to natural language reasoning, LMs still face challenges such as hallucination, generating incorrect intermediate reasoning steps, and making mathematical errors. Recent research has focused on enhancing LMs through self-improvement using feedback. Nevertheless, existing approaches relying on a single generic feedback source fail to address the diverse error types found in LM-generated reasoning chains. In this work, we propose Multi-Aspect Feedback, an iterative refinement framework that integrates multiple feedback modules, including frozen LMs and external tools, each focusing on a specific error category. Our experimental results demonstrate the efficacy of our approach to addressing several errors in the LM-generated reasoning chain and thus improving the overall performance of an LM in several reasoning tasks. We see a relative improvement of up to 20% in Mathematical Reasoning and up to 18% in Logical Entailment.

Deepak Nathani, Lovish Madaan, Nicholas Roberts et al.

We introduce Meta MLGym and MLGym-Bench, a new framework and benchmark for evaluating and developing LLM agents on AI research tasks. This is the first Gym environment for machine learning (ML) tasks, enabling research on reinforcement learning (RL) algorithms for training such agents. MLGym-bench consists of 13 diverse and open-ended AI research tasks from diverse domains such as computer vision, natural language processing, reinforcement learning, and game theory. Solving these tasks requires real-world AI research skills such as generating new ideas and hypotheses, creating and processing data, implementing ML methods, training models, running experiments, analyzing the results, and iterating through this process to improve on a given task. We evaluate a number of frontier large language models (LLMs) on our benchmarks such as Claude-3.5-Sonnet, Llama-3.1 405B, GPT-4o, o1-preview, and Gemini-1.5 Pro. Our MLGym framework makes it easy to add new tasks, integrate and evaluate models or agents, generate synthetic data at scale, as well as develop new learning algorithms for training agents on AI research tasks. We find that current frontier models can improve on the given baselines, usually by finding better hyperparameters, but do not generate novel hypotheses, algorithms, architectures, or substantial improvements. We open-source our framework and benchmark to facilitate future research in advancing the AI research capabilities of LLM agents.

Zhaotian Weng, Antonis Antoniades, Deepak Nathani et al.

Open-ended self-improving agents can autonomously modify their own structural designs to advance their capabilities and overcome the limits of pre-defined architectures, thus reducing reliance on human intervention. We introduce Group-Evolving Agents (GEA), a new paradigm for open-ended self-improvements, which treats a group of agents as the fundamental evolutionary unit, enabling explicit experience sharing and reuse within the group throughout evolution. Unlike existing open-ended self-evolving paradigms that adopt tree-structured evolution, GEA overcomes the limitation of inefficient utilization of exploratory diversity caused by isolated evolutionary branches. We evaluate GEA on challenging coding benchmarks, where it significantly outperforms state-of-the-art self-evolving methods (71.0% vs. 56.7% on SWE-bench Verified, 88.3% vs. 68.3% on Polyglot) and matches or exceeds top human-designed agent frameworks (71.8% and 52.0% on two benchmarks, respectively). Analysis reveals that GEA more effectively converts early-stage exploratory diversity into sustained, long-term progress, achieving stronger performance under the same number of evolved agents. Furthermore, GEA exhibits consistent transferability across different coding models and greater robustness, fixing framework-level bugs in 1.4 iterations on average, versus 5 for self-evolving methods.

Deepak Nathani, Cheng Zhang, Chang Huan et al.

Proactive agents that anticipate user needs and autonomously execute tasks hold great promise as digital assistants, yet the lack of realistic user simulation frameworks hinders their development. Existing approaches model apps as flat tool-calling APIs, failing to capture the stateful and sequential nature of user interaction in digital environments and making realistic user simulation infeasible. We introduce Proactive Agent Research Environment (Pare), a framework for building and evaluating proactive agents in digital environments. Pare models applications as finite state machines with stateful navigation and state-dependent action space for the user simulator, enabling active user simulation. Building on this foundation, we present Pare-Bench, a benchmark of 143 diverse tasks spanning communication, productivity, scheduling, and lifestyle apps, designed to test context observation, goal inference, intervention timing, and multi-app orchestration.

Gurusha Juneja, Deepak Nathani, William Yang Wang

Process Reward Models (PRMs) enhance reasoning ability of LLMs by providing step-level supervision. However, their widespread adoption is limited due to expensive manual step-level annotation and poor generalization of static training data to novel errors. We introduce Adversarially Trained PRMs (\texttt{APRM}), where a Generator ($G$) learns to produce reasoning errors to deceive a PRM ($R$), while $R$ concurrently learns to detect them. This interaction yields progressively harder negatives for $R$, improving its robustness and generalization to novel errors without requiring manual step-level labels. Averaged across diverse mathematical reasoning benchmarks, \texttt{APRM} improves solver accuracy by $+3.4$ percentage points (pp) over the strongest PRM baseline. \texttt{APRM} achieves gains of $+5.3$ pp on out-of-distribution tasks.

Eshaan Tanwar, Deepak Nathani, William Yang Wang et al.

Large Language Models (LLMs) fine-tuned for specific domains exhibit strong performance; however, the underlying mechanisms by which this fine-tuning reshapes their parametric space are not well understood. Prior works primarily focus on auto-regressive or general-purpose instruct models, leaving domain-specialised LLMs under-explored. We present the first systematic study of domain-specific fine-tuning in large medical language models. Our analysis reveals that fine-tuning modifies only a small subset of the representational subspace, essentially preserving the pre-trained model's representation. To interpret these changes in subspaces, we propose tuning vectors, a novel framework inspired by task vectors, which explicitly capture the directional parameter shifts induced by fine-tuning. We demonstrate that these vectors are critical for enhancing both instruction-following and generation quality. Furthermore, combining tuning vectors across different domains yields improved generalisation. Upon closer inspection of directional alignment, we find these vectors primarily write new directional information into the MLP layers of the model, while amplifying existing directions in attention heads. Our findings offer new insights into LLM adaptation and provide a general, interpretable framework for analysing specialisation in large language models.

Kalpesh Krishna, Deepak Nathani, Xavier Garcia et al.

Style transfer is the task of rewriting a sentence into a target style while approximately preserving content. While most prior literature assumes access to a large style-labelled corpus, recent work (Riley et al. 2021) has attempted "few-shot" style transfer using only 3-10 sentences at inference for style extraction. In this work we study a relevant low-resource setting: style transfer for languages where no style-labelled corpora are available. We notice that existing few-shot methods perform this task poorly, often copying inputs verbatim. We push the state-of-the-art for few-shot style transfer with a new method modeling the stylistic difference between paraphrases. When compared to prior work, our model achieves 2-3x better performance in formality transfer and code-mixing addition across seven languages. Moreover, our method is better at controlling the style transfer magnitude using an input scalar knob. We report promising qualitative results for several attribute transfer tasks (sentiment transfer, simplification, gender neutralization, text anonymization) all without retraining the model. Finally, we find model evaluation to be difficult due to the lack of datasets and metrics for many languages. To facilitate future research we crowdsource formality annotations for 4000 sentence pairs in four Indic languages, and use this data to design our automatic evaluations.

Jatin Chauhan, Deepak Nathani, Manohar Kaul

We propose to study the problem of few shot graph classification in graph neural networks (GNNs) to recognize unseen classes, given limited labeled graph examples. Despite several interesting GNN variants being proposed recently for node and graph classification tasks, when faced with scarce labeled examples in the few shot setting, these GNNs exhibit significant loss in classification performance. Here, we present an approach where a probability measure is assigned to each graph based on the spectrum of the graphs normalized Laplacian. This enables us to accordingly cluster the graph base labels associated with each graph into super classes, where the Lp Wasserstein distance serves as our underlying distance metric. Subsequently, a super graph constructed based on the super classes is then fed to our proposed GNN framework which exploits the latent inter class relationships made explicit by the super graph to achieve better class label separation among the graphs. We conduct exhaustive empirical evaluations of our proposed method and show that it outperforms both the adaptation of state of the art graph classification methods to few shot scenario and our naive baseline GNNs. Additionally, we also extend and study the behavior of our method to semi supervised and active learning scenarios.

Charu Sharma, Deepak Nathani, Manohar Kaul

We present an alternate formulation of the partial assignment problem as matching random clique complexes, that are higher-order analogues of random graphs, designed to provide a set of invariants that better detect higher-order structure. The proposed method creates random clique adjacency matrices for each k-skeleton of the random clique complexes and matches them, taking into account each point as the affine combination of its geometric neighbourhood. We justify our solution theoretically, by analyzing the runtime and storage complexity of our algorithm along with the asymptotic behaviour of the quadratic assignment problem (QAP) that is associated with the underlying random clique adjacency matrices. Experiments on both synthetic and real-world datasets, containing severe occlusions and distortions, provide insight into the accuracy, efficiency, and robustness of our approach. We outperform diverse matching algorithms by a significant margin.

Deepak Nathani, Jatin Chauhan, Charu Sharma et al.

The recent proliferation of knowledge graphs (KGs) coupled with incomplete or partial information, in the form of missing relations (links) between entities, has fueled a lot of research on knowledge base completion (also known as relation prediction). Several recent works suggest that convolutional neural network (CNN) based models generate richer and more expressive feature embeddings and hence also perform well on relation prediction. However, we observe that these KG embeddings treat triples independently and thus fail to cover the complex and hidden information that is inherently implicit in the local neighborhood surrounding a triple. To this effect, our paper proposes a novel attention based feature embedding that captures both entity and relation features in any given entity's neighborhood. Additionally, we also encapsulate relation clusters and multihop relations in our model. Our empirical study offers insights into the efficacy of our attention based model and we show marked performance gains in comparison to state of the art methods on all datasets.