arXiv subjects

Jesse Clark

Publications and source records attributed to Jesse Clark.

6 recordsLinked to original sources

arXiv2026

Outcome-Routed Distillation and Reinforcement Learning for Diffusion Language Models

On-policy distillation provides dense supervision on the student's own samples, but its targets still come from the teacher. Verifier reinforcement learning (RL) instead uses an outcome signal, but it provides no direct constraint on capabilities outside the verifier. Combining the two is difficult for diffusion language models (dLLMs), because bidirectional denoising makes the policy likelihoods required by ratio-based RL unreliable. We develop a likelihood-free pipeline for a mixture-of-experts dLLM. Each rollout group is routed by verifier outcome. Mixed groups use ratio-free, group-centred REINFORCE. If every student rollout fails but a teacher succeeds, the model receives dense KL supervision on the student's denoising canvases. All other groups are skipped. A KL term anchored to the distilled checkpoint limits changes outside the verifier reward. The tool-calling teacher succeeds on only 18.6% of the training prompts. At 24 denoising steps, distillation exceeds the teacher by 6.77 percentage points, and routing improves the distilled checkpoint by a further 18.31 points. Routing lowers reasoning-trace accuracy by 0.71 points but raises direct-answer accuracy on the same items by 13.21 points. The anchored arm restores trace accuracy. Its tool-calling score differs by -0.16 points, with a confidence interval that includes zero, although this comparison also removes a small fallback distillation route. A domain-conditional recipe retains a 16.37-point tool-calling gain; a second seed gives 18.23 points and reproduces both preservation results, but the math result is seed-dependent. We pre-registered every arm and evaluated each comparison on paired items in a single sweep on one machine.

cs.LG↗

arXiv2026

Context-Tower Conversion Preserves Generation While Freezing Retains Knowledge: Low-Budget AR-to-Diffusion Conversion of MoE LLMs

Converting a pretrained autoregressive (AR) model to a diffusion language model (dLLM) enables parallel generation without pretraining a new model. Published conversion methods differ by roughly three orders of magnitude in training data and have not been compared under a common protocol. We compare two conversions of the same 30B Mixture-of-Experts (MoE) parent, holding the corpus, supervised-token budget, trainable parameter set and evaluation harness fixed, each under its own training recipe. The in-place model updates a subset of the parent's weights using denoising and representation-alignment losses; the frozen-tower model instead conditions through cross-attention on a frozen causal copy of the parent. With 1B training tokens, the frozen-tower model scores 71.60 on HumanEval pass@10 against 6.19 for the in-place model, an 11.6x improvement. At the same budget it also keeps 95% of the parent's GSM8K score and 99% of its MMLU-Pro score. A dense-parent experiment reproduces the HumanEval separation. Within the two-tower design at about 500M tokens, freezing the context tower retains substantially more MMLU-Pro performance than training it, while both give similar observed HumanEval scores. Our theoretical analysis establishes that both conversion classes contain an exact sampler for the AR parent under a hard attention mask and left-to-right commitment of one position per round. Under a shared loss, freezing removes the gradient contribution through the context states. Furthermore, evaluation protocol substantially affects a published 500B-token conversion's scores in both directions across tasks, while its AR parent's scores vary by less than three points, so comparing dLLMs needs a common protocol. These results show that, in the tested low-budget regime, the frozen-tower configuration retains substantially more of the parent's generation performance than in-place conversion.

cs.LG↗

arXiv2025

Generalized Contrastive Learning for Multi-Modal Retrieval and Ranking

Contrastive learning has gained widespread adoption for retrieval tasks due to its minimal requirement for manual annotations. However, popular training frameworks typically learn from binary (positive/negative) relevance, making them ineffective at incorporating desired rankings. As a result, the poor ranking performance of these models forces systems to employ a re-ranker, which increases complexity, maintenance effort and inference time. To address this, we introduce Generalized Contrastive Learning (GCL), a training framework designed to learn from continuous ranking scores beyond binary relevance. GCL encodes both relevance and ranking information into a unified embedding space by applying ranking scores to the loss function. This enables a single-stage retrieval system. In addition, during our research, we identified a lack of public multi-modal datasets that benchmark both retrieval and ranking capabilities. To facilitate this and future research for ranked retrieval, we curated a large-scale MarqoGS-10M dataset using GPT-4 and Google Shopping, providing ranking scores for each of the 10 million query-document pairs. Our results show that GCL achieves a 29.3% increase in NDCG@10 for in-domain evaluations and 6.0% to 10.0% increases for cold-start evaluations compared to the finetuned CLIP baseline with MarqoGS-10M. Additionally, we evaluated GCL offline on a proprietary user interaction data. GCL shows an 11.2% gain for in-domain evaluations. The dataset and the method are available at: https://github.com/marqo-ai/GCL.

cs.IR↗

arXiv2024

Designing Interfaces for Multimodal Vector Search Applications

Multimodal vector search offers a new paradigm for information retrieval by exposing numerous pieces of functionality which are not possible in traditional lexical search engines. While multimodal vector search can be treated as a drop in replacement for these traditional systems, the experience can be significantly enhanced by leveraging the unique capabilities of multimodal search. Central to any information retrieval system is a user who expresses an information need, traditional user interfaces with a single search bar allow users to interact with lexical search systems effectively however are not necessarily optimal for multimodal vector search. In this paper we explore novel capabilities of multimodal vector search applications utilising CLIP models and present implementations and design patterns which better allow users to express their information needs and effectively interact with these systems in an information retrieval context.

cs.IR↗

arXiv2024

The Impacts of Data, Ordering, and Intrinsic Dimensionality on Recall in Hierarchical Navigable Small Worlds

Vector search systems, pivotal in AI applications, often rely on the Hierarchical Navigable Small Worlds (HNSW) algorithm. However, the behaviour of HNSW under real-world scenarios using vectors generated with deep learning models remains under-explored. Existing Approximate Nearest Neighbours (ANN) benchmarks and research typically has an over-reliance on simplistic datasets like MNIST or SIFT1M and fail to reflect the complexity of current use-cases. Our investigation focuses on HNSW's efficacy across a spectrum of datasets, including synthetic vectors tailored to mimic specific intrinsic dimensionalities, widely-used retrieval benchmarks with popular embedding models, and proprietary e-commerce image data with CLIP models. We survey the most popular HNSW vector databases and collate their default parameters to provide a realistic fixed parameterisation for the duration of the paper. We discover that the recall of approximate HNSW search, in comparison to exact K Nearest Neighbours (KNN) search, is linked to the vector space's intrinsic dimensionality and significantly influenced by the data insertion sequence. Our methodology highlights how insertion order, informed by measurable properties such as the pointwise Local Intrinsic Dimensionality (LID) or known categories, can shift recall by up to 12 percentage points. We also observe that running popular benchmark datasets with HNSW instead of KNN can shift rankings by up to three positions for some models. This work underscores the need for more nuanced benchmarks and design considerations in developing robust vector search systems using approximate vector search algorithms. This study presents a number of scenarios with varying real world applicability which aim to better increase understanding and future development of ANN algorithms and embedding

cs.IR↗

arXiv2018

Fog Robotics for Efficient, Fluent and Robust Human-Robot Interaction

Active communication between robots and humans is essential for effective human-robot interaction. To accomplish this objective, Cloud Robotics (CR) was introduced to make robots enhance their capabilities. It enables robots to perform extensive computations in the cloud by sharing their outcomes. Outcomes include maps, images, processing power, data, activities, and other robot resources. But due to the colossal growth of data and traffic, CR suffers from serious latency issues. Therefore, it is unlikely to scale a large number of robots particularly in human-robot interaction scenarios, where responsiveness is paramount. Furthermore, other issues related to security such as privacy breaches and ransomware attacks can increase. To address these problems, in this paper, we have envisioned the next generation of social robotic architectures based on Fog Robotics (FR) that inherits the strengths of Fog Computing to augment the future social robotic systems. These new architectures can escalate the dexterity of robots by shoving the data closer to the robot. Additionally, they can ensure that human-robot interaction is more responsive by resolving the problems of CR. Moreover, experimental results are further discussed by considering a scenario of FR and latency as a primary factor comparing to CR models.

cs.RO↗