Skip to main content
Search

Seoul National University

33 related articles

Subscribe to updates
  1. arXiv

    Inverse rendering stops bolting priors on as losses: shared lattice variables lift relighting PSNR by 1.5–2.5 dB

    Physically based inverse rendering must separate materials from illumination, and the decomposition is ambiguous. Learned material predictions are usually added as auxiliary regularization losses, which change the objective and require rebalancing against photometric fitting. This work instead restricts the optimization space: multi-view material predictions are fused onto a mesh, a permutohedral lattice is built in the prior feature space, and surface points with similar predictions share material variables while the objective stays purely photometric. On Synthetic4Relight and TensoIR-Synthetic, relighting PSNR with two predictors is 1.5–2.5 dB above the strongest evaluated mesh-based baseline, and it beats a regularization loss using the same prior in controlled comparisons.

  2. arXiv

    Full-duplex speech models collapse around ten minutes: self-distillation plus two-resolution context cuts content and voice degradation by 71–96%

    Full-duplex spoken dialogue models drift over long sessions, becoming slower to respond, less coherent and less consistent in voice, and long-session stability had not been targeted by dedicated training or architecture. A closed-loop simulated user quantifies this minute by minute across four public models, then privileged-context self-distillation applies lightweight LoRA adaptation and a two-resolution context (TRC) compresses older history. Across three architecturally distinct models, self-distillation mitigates degradation within the ten-minute training horizon, and adding TRC reduces spoken-content and voice degradation by 71–96% relative to baseline while extending the gains to twenty-minute sessions.

  3. arXiv

    Diffusion models move temporal structure into the noising process: DSSD lowers forecast error on weather fields and robot videos

    Diffusion models for spatiotemporal prediction usually add independent noise to each future frame and leave temporal dependence to the denoiser. DSSD instead builds the temporal operator of the corruption process from a learnable diagonal state-space model whose decay rates and frequencies train jointly with the denoiser, and uses the Cholesky factor of that model's stationary covariance so the operator stays causal, exactly invertible and unit-variance per frame. On SEVIR, Turbulence, BAIR and RoboNet it beats diffusion baselines sharing the same denoiser, with raw CRPS 3.975 versus 4.174 for the strongest SEVIR baseline, DPM.

  4. arXiv

    JSSA makes token-level sparse decoding gather-free, reaching up to full-KV throughput at 128K context

    Long-context decoding is bottlenecked by repeatedly reading a growing KV cache, and sparse attention reads only part of it, yet token-level selection has not reliably delivered both accuracy and lower latency. JSSA scores candidate tokens in a shared low-dimensional query–key subspace and pairs it with a gather-free indexed-attention kernel that keeps a distinct selected token set for each KV head. Across GQA, hybrid sliding-window/global attention and MLA models it stays closest to full-KV accuracy, and on a B200 at 128K context its decoding throughput reaches up to that of full-KV attention.

  5. medRxiv

    One clinical foundation model, two ways to predict: representations win on risk, generation wins on same-day orders

    FERMAT, a 65-million-parameter clinical foundation model pretrained on records from 3.77 million patients, was compared against itself on the same patients: patient representations predicted five-year kidney, fatty liver, and diabetes risk better, while generated trajectories predicted the diagnoses, prescriptions, and procedures recorded together on the next clinical date better.

  6. arXiv

    VLA action chunks get a validity check before execution: PAV predicts each prefix's outcome and lifts LIBERO average success to 94.65%

    Vision-language-action (VLA) policies emit a whole action sequence per query, but how much of it to execute was previously judged only indirectly, through generation stability or prediction agreement. PAV instead predicts, inside the policy's own task-conditioned representation, what a candidate prefix will lead to, and a verifier trained on trajectory-level outcomes scores whether the task can still be completed. On the four LIBERO suites, PAV averages 94.65% success at 14.0 policy queries per episode, above direct verification without future prediction (84.55%, 22.3 queries) and prediction in DINOv2 features (87.00%, 22.2 queries).

  7. medRxiv

    Three LLMs locked predictions before ESC 2026: from registered design alone, the best AUC for separating positive from negative trials was 0.80

    Before the European Society of Cardiology Congress 2026 opened, three large language models locked predictions of whether each trial's primary endpoint would be met, using only design information public at registration. Of 48 scorable trials, 28 were positive; Claude, GPT and Gemini separated positive from negative trials with AUCs of 0.80, 0.79 and 0.74, all above chance. Among 26 trials whose effect was reported exactly as locked, the direction was correct in 88% of predictions, and the median multiplicative error of ratio estimates was about 15%.

  8. bioRxiv

    Spatial transcriptomics skips cell segmentation first: SSAF-T types cells from a literature knowledge graph, then uses that map to guide segmentation

    Spatial transcriptomics analysis has long depended on cell segmentation, whose boundaries and transcript assignments are error-prone; segmentation-free methods avoid that step but often need an external single-cell reference or a preset number of cell types. SSAF-T chains segmentation-free cell typing to segmentation: the agent CeLLMan labels clusters from a literature-grounded knowledge graph and marker genes, the labels are spread across the tissue map, and that map constrains nucleus-seeded watershed segmentation. On public data including Visium HD mouse small intestine, Stereo-seq mouse brain, Xenium mouse kidney and MERFISH mouse liver, cell-type distributions match tissue anatomy, segmentation is comparable to Baysor and Cellpose, and runs are faster.

  9. arXiv

    MLP linearization moves to low-degree rational fits: Qwen3-8B drops 14% of parameters and decodes faster

    Instead of relying only on polynomial random features, MLP linearization now approximates activations with low-degree rational functions and splits each layer, via partial fractions, into separable weight-side and input-side inner products. Across synthetic fitting, vision, 3D and language models this route gives lower error at the same feature budget; compressing Qwen3-8B by replacing its last eight feed-forward layers removes 1.17B parameters (14% of the model) and raises decoding throughput from 763.3 to 809.6 tokens/s, with reasoning and knowledge benchmarks roughly at parity with the uncompressed model.

  10. Advanced Science

    Metal sites no longer need high-resolution crystals: prune-then-optimize predicts location and ion type on cryo-EM and AlphaFold models

    PrOMet splits metal-site prediction into a pruning step and an optimization step, reaching a PR-AUC of 0.739 for site localization on a held-out test set while naming the bound metal type from the same structure.