Skip to main content
Search

University of Waterloo

12 related articles

Subscribe to updates
  1. arXiv

    EmoReward stops emotional video RL from trusting fixed rewards: last run's traces set the next run's weights

    Emotional video generation has no single correct target, and fixed fusion of visual, motion and text-alignment rewards captures emotional intent poorly. EmoReward distills reward traces from one completed training run into per-emotion-regime reliability estimates, converts them into bounded weight multipliers, and combines them with current-video instance gates in the next run without updating any reward-model parameters. On HunyuanVideo 1.5 with DanceGRPO, FlowGRPO and SAGE-GRPO, VideoAlign Overall rises under all three algorithms, including SAGE-GRPO with EmoReward moving from -0.8269 to -0.2624.

  2. arXiv

    Checkpoints stop costing bits: reference exponents predict displacement length, cutting storage up to 27%

    ULPack encodes weight displacements between adjacent checkpoints as ordered-integer residuals and conditions their entropy models on the reference weight's exponent, cutting storage by up to 27% against ZipNN across bfloat16, float16, and float32 trajectories.

  3. arXiv

    RucTangle splits coding-agent patches into commit histories that stay runnable at every commit and lifts pass@1 by 16.6% on average in regression repair

    The work introduces RucTangle, an agentic commit-untangling method that groups a patch's changes by development purpose, orders them, applies each commit and checks whether the code still runs, revising order first, then regrouping via the agent, and finally merging adjacent commits while preserving the patch's final code; on 131 agent-generated patches it is the only method that reconstructs every patch with all commits runnable, at a median of eight commits, whereas baselines leave 20.6%–37.4% of histories with at least one unrunnable version. The authors also present TangleEval, which has two coding agents repair 453 regression-introducing patches; adding RucTangle histories plus git bisect results raises pass@1 from 26.9% to 31.3% and from 30.8% to 36.0%.

  4. arXiv

    Project Greenhouse pre-trained a 3B model from scratch on 8 H100s and fine-tuned it into a pointwise reranker that beats comparable open-weight baselines on TREC DL and BEIR, trailing only GPT-6.1 Sol

    Project Greenhouse models LLM training as a directed property hypergraph and uses it to define what makes a model fully open and sovereign; as a first milestone, the authors pre-train a 3.29B-parameter causal language model from scratch on ClimbMix using only publicly available datasets, then fine-tune it on RLHN-250K with localized contrastive estimation to obtain the pointwise reranker Gaggle (base reranker). Its mean nDCG@10 exceeds all compared fine-tuned pointwise and listwise rerankers on TREC DL 2019–2023 and seven BEIR collections, with only GPT-6.1 Sol scoring higher on TREC DL.

  5. arXiv

    Sequential resetting procedures deliver horizon-independent FDR control for LLM watermark detection and financial backtesting

    The authors propose sequential resetting procedures built on e-values and test supermartingales that reset the capital to one after each rejection and continue testing, thereby reporting multiple intervals containing null violations; a refinement rule reports only observations after the within-block minimum of the capital. Under independent null e-values the FDR bound is α, while under dependence between null and non-null data the bound is α times a logarithmic factor with anytime validity, demonstrated on LLM watermark detection and ES backtesting.

  6. arXiv

    AI Theorist autonomously builds an α-RuCl₃ exciton model from first-principles and many-body calculations, explaining unpublished optical and photocurrent observations

    The authors introduce AI Theorist, a system of AI agents for autonomous discovery of physical models through hypothesis generation, first-principles calculations, and evidence-driven refinement, and apply it to the Kitaev quantum spin liquid candidate α-RuCl₃, producing a new interpretation of optical and photocurrent observations that identifies distinct excitonic states with contrasting optical selection rules and real-space distributions.

  7. arXiv

    CSIR planning framework lets robots find target people under incomplete information, beating distance-only graph baselines while semantics alone destabilizes search

    For the Person Goal Navigation (PersonNav) problem, the work proposes a planning framework that combines distance and semantic information about the recipient (habits and intent) weighted by the trust of user-provided information, and builds a synthetic benchmark of scenarios with actors, items, and requests to evaluate performance before real-world deployment while simulating natural-language human-robot interaction; results show the informed search outperforms classical distance-based graph baselines, semantics alone leads to ungrounded and sporadic search, an LLM-based variant performs comparably, and hardware tests show the method supports real-world embodiment.

  8. arXiv

    Clean Cuts Second-Order Optimizer Memory to Linear via Nyström Sketching, Enabling 13B Pre-training on a Single 80GB GPU

    The authors introduce Clean and its low-precision variant Q-Clean: randomized Nyström approximations replace SOAP's explicit left and right preconditioners, reducing optimizer-state memory from quadratic to linear in layer dimensions, while four orthogonal blocks (Main, A, B, C) reintegrate off-subspace components into the update; pre-training LLaMA-350M and 1.3B on C4, Clean reaches perplexity close to SOAP with less optimizer memory than AdamW, Q-Clean cuts optimizer memory by over 50% versus Muon, Clean reaches AdamW's final performance 26% faster in wall-clock time, and the methods uniquely enable pre-training a 13B model on a single 80GB H100.

  9. arXiv

    X-Tree mines reusable skills from trajectories into a tree, lifting success by up to 4.5%, 5.8% and 4.1% on WebArena, ScienceWorld and WebShop at matched data and budget

    The work introduces X-Tree: raw actions are first canonicalized into typed tokens such as type⟨date⟩ or click⟨button⟩, then adjacent pairs are recursively merged under an X-Score that weighs recurrence, span length and the fraction of occurrences in successful episodes, yielding a deterministic reusable experience tree mined from a trajectory pool with zero LLM calls; X-Tree is then integrated into three training settings — offline RL with each node as a training instance under a step-matching reward plus a depth-scaled completion bonus, online RLVR with an adaptive skill bonus that supplies signal while the verifier is silent and anneals as it becomes informative, and on-policy self-distillation where a rendered X-Tree replaces the LLM-written skill bank as the self-teacher's privileged c

  10. arXiv

    SCOPD lifts VLM performance at 10% visual-token retention from 86.37% to 92.43% via sparse-context on-policy self-distillation

    Using a fixed-context Pass@K analysis, this work shows that degradation after visual-token pruning is not only lost information but also unreliable use of surviving evidence (a representation-utilization gap), and introduces SCOPD and SCOPD+ sparse-context on-policy self-distillation, raising the normalized average over 13 benchmarks at 10% visual-token retention from 86.37% (Vanilla) to 90.49% and 92.43% without architectural changes, extra inference-time computation, or ground-truth answers.