Inference
-
Towards Robust Vision-Language Reasoning: A Multi-Stage Optimization Pipeline with Masked Diffusion Model
2026-11-11 09:30:15Shunming Liu, Pei Huang, Yajing Sun, Huaxiao LiuAbstract
Discrete diffusion generation is emerging as a competitive alternative to autoregressive generation by denoising the entire token sequence through iterative refinement. Its global planning and progressive correction capabilities are particularly well suited to autonomous driving. However, transferring this generation paradigm from general-purpose settings to driving introduces substantial domain gaps in both training and inference. To unlock its potential, we propose a multi-stage optimization pipeline for standard vision-language model architectures, without auxiliary action experts or specialized mixture-of-experts Transformer, spanning post-training and inference. During supervised fine-tuning, we optimize the representation of planning inputs to suppress hallucinations in task responses. During reinforcement learning, we introduce metric-guided diversity filtering to identify stronger learning starting points and improve policy exploration. At inference time, we propose Dynamic Autonomous Driving Decoding (DAD), which adaptively adjusts decoding modes according to scene complexity. Based on this pipeline, we develop Masked Diffusion Drive (MD-Drive), improving its performance by 29.7\% over the unoptimized model. Experiments on the NAVSIM closed loop benchmark and the nuScenes and DriveCoT open loop datasets show that MD-Drive outperforms autoregressive VLM baselines with the same architecture, while achieving a more favorable balance between performance and computational cost than systems with different architectures through discrete diffusion training and inference.
Abstract (translated)
URL
https://arxiv.org/abs/2611.15238
PDF
-
WordInferBench: Can Agents Infer How to Edit from Word Document Context?
2026-11-11 08:36:16Jiayu Ding, Yupan Huang, Jinyoung Park, Yilin Jia, Wenshan Wu, Tengchao Lv, Shaohan Huang, Yuzhong Zhao, Xiangyang Zhou, Tao Ge, Sun Mao, Xun Wang, Huitian Jiao, Si-Qing Chen, Lei Cui, Nanning Zheng, Furu WeiAbstract
Document editing is among the most common knowledge-work tasks, yet existing benchmarks for evaluating autonomous agents on Word document editing provide fully specified attribute-modification instructions. They do not evaluate agents' ability to infer the appropriate attributes from document context, a capability essential in real-world editing where users expect consistency without spelling out every attribute. We introduce WordInferBench, a benchmark comprising 280 base document-editing instances, each paired with three instruction conditions: descriptive (the instruction explicitly lists every attribute change), referential (the instruction points to another object as an exemplar), and inferential (the agent must examine document context to infer what attributes to apply). To evaluate contextual attribute inference, each base instance is constructed so that the target formatting is recoverable from an in-document exemplar. The instances are derived from real Word documents collected from Common Crawl and cover 13 object types and 81 attributes. We evaluate six recent coding-agent systems and find that the best-performing evaluated system, Claude Code with Opus 4.6, achieves a 62.9% task-level pass rate on descriptive variants but only 40.7% on inferential variants, indicating that current systems complete tasks less reliably when the required formatting changes must be determined from document context.
Abstract (translated)
URL
https://arxiv.org/abs/2611.15189
PDF
-
Reframing Video Inverse Tone Mapping as Blind Posterior Inference via Latent Luminance Supervision
2026-11-11 07:36:06Hanwool Sul, Chang-Hoon Jeong, Kisung Shin, Bomseok Park, Donghoon Cha, Hansl von Kwon, DoHyung Ku, Byoung-Tak ZhangAbstract
High-dynamic-range (HDR) displays are now common, but most video content, such as archival footage, broadcast material, and the output of generative video models, exists only in standard dynamic range (SDR). The need is sharpest for generated video, which has no HDR master at all and is graded alongside live-action and visual-effects material finished in HDR. Converting SDR to HDR means undoing a tone-mapping operation that was never recorded, and clipping destroys the radiance wherever it acts, so HDR recovery is a blind inverse problem. Methods built on generative video priors fill clipped regions with plausible structure, but their reconstructions are systematically biased: in every public method we test, saturated content leans too bright or too dark, and the direction of the lean moves with the tone operator behind the input. We call this failure the luminance error, measure both its magnitude and its sign, and make it the focus of this work. Reframing ITM as posterior inference under an unknown operator locates the failure in two places and yields two corrections. A fixed affine functional of the video-diffusion latent reads log-luminance without decoding; on that coordinate an asymmetric loss selects an expectile of the posterior, applied only where clipping leaves luminance unconstrained. We also synthesize training SDR through an explicit family of tone operators, so the model marginalizes over what the observation cannot show. Trained this way, our luminance error stays flatter across operators and smaller on average than any public method, and the full system achieves the best perceptual quality on all three benchmarks.
Abstract (translated)
URL
https://arxiv.org/abs/2611.15146
PDF
-
Read Once, Answer Many: Parallel Prompting for Shared-Context LLM Inference
Abstract
LLM workloads sometimes ask many independent questions about the same long document, as in document QA, data labeling and LLM-based evaluation. Prefix caching removes the redundant prefill shared by these queries, but each decode step still reads the document's key-value (KV) cache once per question, so decoding is bound by memory bandwidth and scales with the number of questions. We present Parallel Prompting, which decodes P questions as P independent streams of a single request all at once, fixing this issue. In our method, every step reads the shared document once for all P streams, and a stream-id attention mask, evaluated inside the attention kernel, keeps each answer conditioned only on the document and its own question. We implement Parallel Prompting in SGLang by adding about 80 lines to three existing files and a small engine package; the mask needs no dense tensor, and every decode step has a fixed shape that is captured as a CUDA graph. With CodeLlama-7B, 8k-16k-token documents and 64-256 questions per document, Parallel Prompting is 3.7x faster than stock SGLang with its FlashInfer backend on an A100 and 5.9-6.0x faster on an RTX A6000, on two SGLang releases, including the most recent release. Timed on identical inputs, it is faster than stock SGLang in all 144 runs (2 GPUs x 2 releases x 6 workloads x 6 input sets). Code: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2611.15130
PDF
-
Expanding the Decision Space of Pretrained Models through Structured Re-Decoding
2026-11-11 06:37:08Tinglan Wang, Qun Su, Shihang Wang, Chi Wang, Lin Wang, Shaobo Chen, Haixu Liu, Hai Wei, Yin Ma, Yu Kang, Qi Gao, Xiaojun Yao, Tingjun Hou, Kai WenAbstract
Native decoders limit how pretrained models combine local predictions under new structural constraints. We introduce Structured Re-Decoding (SRD), a training-free framework that constructs local candidates from model scores, samples, or external specifications, jointly optimizes their selection, and verifies reconstructed outputs. We show that candidate recombination expands feasible support and can improve attainable reward under a fixed KL divergence budget. Across molecular generation, protein design, RNA structure prediction, and antibody modeling, SRD enables improved structured inference, including verified molecules absent from the original sample pool. External chemical templates further extend SRD to noncanonical amino acids, peptide cyclization, and grammar-guided molecular construction. Preliminary experiments on a physical coherent Ising machine demonstrate the feasibility of hardware-based optimization, suggesting broader opportunities for quantum and quantum-inspired approaches to structured inference and combinatorial optimization across scientific domains.
Abstract (translated)
URL
https://arxiv.org/abs/2611.15109
PDF
-
CFGComposer: Compositional Audio-Text-to-Motion Generation via Copula Score Correction
2026-11-11 06:28:18Pengyang Shi, Jinlu Zhang, Chunze Lin, Yizhou Wang, Li YiAbstract
Generating whole-body gestures that are simultaneously rhythmically synchronized with speech audio and semantically aligned with text descriptions is a fundamental challenge in embodied AI and social robotics. We tackle this Audio-Text-to-Motion (AT2M) problem by composing two independently-trained diffusion models at inference time and refine bias in joint inference by modeling the copula score $s_I$. A rigorous theoretical analysis reveals that naive epsilon blending corresponds to a Product-of-Experts formulation that omits the copula score $s_I$: the cross-modal interaction between audio and text that neither marginal model can capture. We propose CFGComposer: a jointly-trained AT2M diffusion model that estimates $s_I$ via a CFG analogy and injects it as an $\epsilon$-space correction. Experiments on BEAT demonstrate best-in-class semantic fidelity (SRGR $0.9916$, FID $27.18$) with GT-level beat alignment (BA $0.1826$ vs. GT $0.1743$), and 76-83% win rate in blind pairwise human evaluation against all prior methods. Physical deployment on a Unitree G1 humanoid confirms direct sim-to-real transfer.
Abstract (translated)
URL
https://arxiv.org/abs/2611.15101
PDF
-
The Posterior-Prior Epistemic Gap: When World Model Imagination Fails to Carry Task-Relevant Signals
2026-11-11 03:02:55Jaerock Kwon, Elahe Delavari, Donghyun Kim, Haewoon NamAbstract
Model-based planners that roll out imagined futures assume the transition model faithfully propagates task-relevant signals---a property rarely measured directly. In a single-seed case study of a Recurrent State Space Model world model with active-inference planning in the CARLA driving simulator, we identify a \textbf{posterior-prior gap}---posterior (observation-filtered) states retain obstacle information that imagined rollout states progressively lose: a supervised obstacle probe achieves AUC 0.85 on posterior states, but the same probe applied to imagined prior states shows monotonic decay (AUC 0.85$\rightarrow$0.71 over 15 steps; logit separation drops 57\%). Because the probe is identical, the decay isolates the transition model---not the representation---as the information bottleneck. Obstacle avoidance drops to 0\% without privileged runtime coordinates, and the planner's own ensemble-based epistemic term does not flag the loss---ensemble disagreement yields AUC 0.52 as an obstacle classifier (chance level), making the failure an unflagged imagination failure---an unknown unknown at planning time. As a secondary contribution, we show that \textbf{task-conditioned weighting of Effective Free Energy components} over the same world model yields lane keeping ($81.2 \pm 1.8$\% route completion) and obstacle avoidance (28/28 avoided, \emph{using privileged coordinates}) in the CARLA simulator, with runtime weight modulation enabling post-evasion centering.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14955
PDF
-
Emergent Vision-Shaped Thinking in Text-Only Reasoning through Multimodal Training
2026-11-11 02:44:10Qi Zhang, Bohan Wang, Qianru Sun, Hanwang ZhangAbstract
Does multimodal training shape how a language model thinks, even when the input is text only? We test this with targeted causal ablations across 10 multimodal configurations from 6 model families, with matched text-only backbones and cross-family transfer controls. We contrast spatial-imagery tasks that invite mental scene construction with symbol-manipulation tasks solvable by symbolic rules alone. We identify a vision-language axis in hidden-state space and ablate its projection at inference. In Chameleon-7B, this ablation drops accuracy on BIG-Bench Hard (BBH) navigate by 27.6 pp while leaving boolean expressions intact (+5.6 pp), a 33.2 pp double dissociation. The effect is direction-specific and layer-localized, extends to mental-simulation tasks such as mental rotation, paper folding, and perspective taking, and is not explained by task difficulty, reasoning depth, or sequential update load. Together, these results reveal emergent vision-shaped thinking (EVT): text-only reasoning that causally recruits representations introduced or strengthened by vision-language training, without assuming an image-like internal format. The measured effect is spatially concentrated at 7-8B and broadens to non-spatial visual attribute binding in the tested 32B configuration. In dual-path unified models (BAGEL, Janus-Pro), the causal handle localizes to the generation pathway as a compact one-dimensional carrier. The understanding pathway also encodes spatial information in a higher-rank, probe-decodable form, but is not the causal handle. To separate decodability from causal use, we analyze concentration, alignment, and semantic location. These diagnostics explain why behavior, probing, and intervention can diverge. Multimodal training thus reshapes the geometry of text-only reasoning, not just how visual input is processed.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14942
PDF
-
Hint-Informed Self-Distillation for Reasoning Segmentation
2026-11-11 02:34:13Kunze Jiang, Yuchen Wang, Kang Li, Qixing ZhangAbstract
Reasoning segmentation requires identifying the target referred to by a query among visually similar objects, yet trajectory-level rewards cannot directly indicate which tokens provide relevant visual evidence. We propose Hint-Informed Self-Distillation (HISD), which uses target-disambiguation evidence for token-level credit assignment. An external model generates Segment Hints offline from training images, queries, and ground-truth target annotations, organizing target identity, discriminative attributes, confusable objects, and anchoring paths. A hint-conditioned exponential moving average teacher scores the student's own trajectories, and HISD computes the teacher-to-student evidence ratio over a token support set to adjust only the magnitudes of positive group-relative advantages, leaving nonpositive advantages unchanged. This design encourages concise reasoning while retaining the information needed for target disambiguation, and the student requires neither hints nor a teacher at inference. Under a unified training protocol, HISD improves ReasonSeg test gIoU and cIoU over our reproduction of DPAD by 3.00 and 3.55 percentage points, respectively. Across the five evaluated splits of ReasonSeg and the RefCOCO family, HISD achieves higher segmentation metrics than this baseline and reduces the average token count of valid reasoning spans by 80.17 to 91.83 percent.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14931
PDF
-
Open Challenges and Opportunities in Human-Agent Systems
2026-11-10 23:55:44Reuth Mirsky, Sarath Sreedharan, Nitay Alon, Stefan Sarkadi, Munindar P. Singh, Liz Sonenberg, Peter Stone, Silvia Tulli, Stylianos Loukas Vasileiou, Shlomo ZilbersteinAbstract
Agentic AI systems increasingly undertake extended tasks and act on people's behalf, making human-agent interaction central to their effectiveness. Successful interaction requires understanding human perspectives, coordinating complementary capabilities, and maintaining productive relationships within socio-technical environments. This paper examines how foundations developed by the Autonomous Agents and Multi-Agent Systems (AAMAS) community inform these demands. We consider interactions ranging from individual human-agent partnerships to teams and larger groups, where participants differ in their roles, decision authority, capabilities, and objectives. We organize computational approaches around the connected challenges of understanding others and deciding how to interact with them. Theory of Mind, belief-desire-intention architectures, goal and plan recognition, inverse planning, and recursive reasoning provide mechanisms for representing and inferring partially observable mental states. Observer-aware planning, model reconciliation, and decision-theoretic communication connect these inferences to action, explanation, and information exchange. Together with shared autonomy, argumentation, and cooperative inverse reinforcement learning, these approaches offer foundations for agentic systems that account for evolving human knowledge, intentions, and expectations. We argue that progress requires treating human interaction as an ongoing decision problem shaped by uncertainty, interdependence, and potentially conflicting objectives.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14846
PDF
-
When Should Semantic Priors Enter Partially-Identified Causal Discovery?
2026-11-10 20:55:34Muhammad Ahmed Mohsin, Saad Bin Ather, Valentin Six, Lancelot Da Costa, Tommaso Salvatori, Bernhard ScholkopfAbstract
Observational causal discovery often leaves causal di- rections unresolved or unstable. Large language mod- els (LLMs) can provide semantic information to resolve these directions, but incorrect proposals can exclude the true causal graph and invalidate downstream inference. We study three separate decisions: where to allocate semantic queries, which proposals to admit, and which causal graphs to retain. Our framework distinguishes population non-identifiability from finite-sample insta- bility, evaluates semantic orientation against statistical and random controls, and examines calibrated admis- sion and error-tolerant DAG retention. Experiments across benchmark networks show that semantic infor- mation improves recovery over coin-flip orientations, but targeting unstable edges has no established ad- vantage over matched-random allocation. The tested admission certificate cannot support nontrivial admis- sion with the available calibration tasks. Moreover, downstream retention depends on reference adequacy: learned references contain the true DAG in only three of 80 planned tasks. These findings separate the value of semantic information from the reliability of its alloca- tion, admission, and retention, and identify calibration size and reference adequacy as key limitations.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14736
PDF
-
Quantitative Assessment of Crystal Structure Prediction
2026-11-10 20:15:46Sergio Rinc\'on, Gabriel Gonz\'alez, Nicol\'as Andrade, Rafael Vel\'asquez, Sebasti\'an Ojeda, Juanita Puentes, Nicol\'as Aparicio, Paula C\'ardenas, Pablo Arbel\'aezAbstract
Crystal Structure Prediction (CSP) aims to predict a material's 3D structure from its chemical composition, with applications from energy storage to biomedical devices. Despite its importance, existing evaluation frameworks suffer from key limitations: they rely on threshold-sensitive or incomparable metrics, focus predominantly on geometric performance while overlooking physical plausibility, and fail to account for polymorphism, where a chemical composition can crystallize into multiple stable structures. In parallel, recent Symmetry-Informed CSP (SICSP) methods often report results alongside standard CSP baselines while receiving crystallographic templates at inference, which constrain the search space and make SICSP a fundamentally different task. To address these limitations, we introduce an Orientation-invariant, Polymorph-Aware, Large Benchmark (OPAL-Bench) that evaluates CSP and SICSP as separate tasks. We evaluate 7 CSP and 2 SICSP methods under OPAL-Bench across 4 datasets using threshold-free geometric and physical distance metrics. Our results show that no method dominates across all metrics, and that geometric performance is not strongly correlated with physical plausibility. We also release OPAL-CSP, an SE(3)-equivariant flow-matching baseline for the CSP track that performs competitively on geometric metrics and ranks among the best on physical ones. By unifying geometric and physical metrics under a polymorph-aware protocol, OPAL-Bench enables a more comprehensive assessment of generative quality in CSP.
Abstract (translated)
URL
https://arxiv.org/abs/2611.10380
PDF
-
Risk-Aware Joint Inference and Resource Control for Physical AI in 6G Networks
2026-11-10 20:01:15Amin Farajzadeh, Melike Erol-KantarciAbstract
Wireless edge inference enables Physical-AI devices to execute edge-inferred actions using transmitted sensory inputs. Since the risk of incorrect actions vary across observations and tasks, the associated uplink transmission of sensory inputs requires different levels of wireless reliability. For instance, applying a fixed signal-to-interference-plus-noise ratio (SINR) requirement can overprovision low-risk tasks while underprovisioning high-risk ones. Therefore, we propose risk-aware control for edge inference (RACE), which maps predicted action-failure probability and task criticality to adaptive SINR targets and combines them with deadline-derived SINR requirements. A slow-timescale large language model (LLM) proposes task requirements for deterministic validation, while a fast hybrid actor selects inference configurations and transmit power levels, complemented by analytical receive beamforming and finite power restoration. Simulations show $11\%$ improvement in the average accuracy-latency-energy objective over fixed-high-SINR control, with near-zero constraint outage.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14696
PDF
-
Together Compiler: Continual Compilation for AI Inference
2026-11-10 19:37:09Reyna Abhyankar, Dylan Lim, Drew Wadsworth, Venkata Hasith Vattikuti, YuTong Julia Zhang, Jung Hwan Heo, Taeksang Kim, Kesavan Ramakrishnan, Evangelene Stanley, Xinyu Fang, Daniel Y. Fu, Simran AroraAbstract
Today's AI workloads are rapidly changing: new models are released more often, model architectures are increasingly diverse, and traffic distributions change frequently. The challenge we face at Together AI is continual optimization for these changing workloads. Compiling a deployment into an optimized forward pass is a challenge that is well-suited for agent hill-climbing. Existing systems struggle with continual compilation due to a tradeoff between measurement throughput and fidelity. Many systems use agents to optimize individual kernels from the inference forward pass using offline benchmarks, but timings are low-fidelity because of data-dependent behavior in primitives like mixture-of-experts (MoE). On the other hand, systems use agents to directly optimize end-to-end engines but suffer from slow engine launch times. To address these challenges, we build Together Compiler (TKCC), the first truly continual agent loop designed for AI inference optimization. To address the throughput-fidelity tradeoff, we propose: (1) a Tiered Persistent Engine that decouples engine launch overheads from forward pass computation; (2) a cost model to select the parallelism configurations that seed the optimization process. We also conduct extensive ablations on agent context management and present our insights on deploying TKCC in production. TKCC outperforms highly-optimized hand-written Megakernels on the Llama 1B workhorse workload by up to 1.7x and the best open source engines on the latest frontier models (Kimi K3, GLM 5.3, MiniMax M3.1 Flash, DeepSeek v4.1 Flash) across hardware platforms by up to 2.6x higher TPS on the InferenceX benchmark. TKCC also sets the state-of-the-art performance at batch size 1, with 1111 TPS on M3.1 Flash and 1205 TPS on DeepSeek v4.1 Flash. Finally, TKCC generalizes beyond frontier language model inference. TKCC serves trillions of tokens across our production fleet.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14681
PDF
-
MITRA-Swarm: Injecting Explicit Linguistic Structure into Foundation Models for Classical Machine Translation
2026-11-10 19:31:07Kayshav Bhardwaj, Pragun Seth, Kush Bhardwaj, Vijay Nagasamy, Kurt KeutzerAbstract
Classical Sanskrit poses three challenges unresolved by single-call monolithic models: morphophonemic fusion Sandhi, dense nominal compounding SamÄsa, and discourse-level zero-anaphora forced by poetic meter. We present MITRA-Swarm, a multi-agent architecture injecting explicit linguistic structure into an otherwise monolithic pipeline: a Sandhi-splitting agent, a parallel array of generalist gemma-4-26B-A4B-it and domain-specialized gemma-2-mitra-it translators, and a discourse-aware Ellipsis Restorer that consults corpus-derived register tags to avoid inventing active-voice subjects in impersonal legal and philosophical registers. Two separated judges evaluate quality: a routing judge ($J_{\text{route}}$, same model family as the generators) driving an internal critique loop, and an independent reporting judge ($J_{\text{report}}=$gpt-4o-mini, no shared vendor lineage) scoring every output exactly once. On a stratified benchmark of 1,000 verses across five registers (23,990 scored verdicts, 12-condition pre-registered ablation), the full swarm reaches $85.90$ MQM against a monolithic baseline of $85.27$ ($\Delta=+0.63$, $p_{\text{adj}}=0.0018$, 95% CI $[+0.30,+0.97]$), and isolated discourse restoration alone accounts for $+0.44$ MQM ($p_{\text{adj}}=0.0288$), outperforming a call-count-matched self-refinement baseline by $+0.51$ MQM ($p_{\text{adj}}=0.0288$). Severe per-verse failures (MQM$<80$) fall from $9.0\%$ to $7.6\%$, at roughly $9\times$ the per-verse inference cost of the baseline. Additionally, $J_{\text{route}}$ scores its own family's outputs $2$-$7$ MQM higher than $J_{\text{report}}$ on every one of the 12 conditions, empirically motivating the two-judge split; and a preflight screen of a 192,151-entry retrieval index against all 12 evaluation works verifies 0 reference leaks across 269 runtime checks. These are small gains on a metric saturating near 85-86.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14675
PDF
-
Thompson Sampling using Prior-fitted Diffusion Transformers
2026-11-10 19:07:33Sihwa Park, Jingsen Zhu, Vinamr Jain, Sheng-Yen Chou, Alexander TereninAbstract
Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. In this work, we introduce \emph{prior-fitted diffusion transformers}, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14656
PDF
-
Mechanistic Interpretability Needs a Logic of Inference
2026-11-10 18:45:07Zhimin Hu, Chen Jiang, Hanbo Xie, N. Apurva Ratan Murty, Sashank VarmaAbstract
Neural networks are increasingly studied not merely as predictive systems, but as complex systems whose internal computations are themselves objects of scientific inquiry. Recent progress in mechanistic interpretability (MI) has made it possible to identify neurons, features, circuits, population structures, and dynamical patterns associated with model behavior. As these tools improve, attention is shifting from what can be observed inside a model to how those observations support mechanistic explanations. We examine the parallel efforts of neuroscience and MI to address this question and identify recurring inferential bottlenecks when inferring mechanisms from complex neural systems. We propose five principles and organize them into a recursive logic for a more rigorous procedure of mechanistic inference. These principles provide a framework that supports scientific discovery in two senses: it helps turn local interpretability findings into cumulative knowledge of mechanisms within models, and it makes those mechanisms sufficiently precise to generate testable hypotheses about domains beyond the models. Evidence beyond MI is ultimately required to establish what model mechanisms reveal about the world.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14483
PDF
-
C$^2$-Routing: Complementary Cropping and Compression for Efficient Vision-Language Inference
2026-11-10 18:32:28Guorui Lu, Qinyu ChenAbstract
Visual Question Answering (VQA) imposes substantial computational overhead on vision-language models (VLMs), motivating model-agnostic input reduction methods that can be readily applied across diverse VLMs, including in black-box settings. Simply enforcing a visual-token cap reduces computation, but it can discard information needed to answer the question. A single reduction strategy also struggles to accommodate different scenarios. Our key observation is that cropping and compression are complementary: cropping highlights small targets and local details, whereas whole-image compression better preserves context for large targets or distributed evidence. Based on this observation, we propose C$^2$-Routing, which adaptively selects cropping or compression according to the input image and question. Since both strategies may still discard key information, we further design a safety gate that uses the predicted joint correctness distribution under original and reduced inputs to guide fallback and protect accuracy. To accommodate consumer GPU resource constraints, all training and evaluation experiments run on RTX 3090 GPUs. To measure gains beyond token capping, we use full-image baselines with the same cap of 1,024 visual tokens. We conduct extensive evaluation on three VLMs and 18 datasets spanning local details, global scenes, scene text, and documents. With the visual budget already capped, C$^2$-Routing further reduces end-to-end FLOPs by 36.53% and visual tokens by 42.36% on average, with a mean relative task-score decrease of only 0.77%.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14469
PDF
-
UGG: Unified Generative Geometry Model
2026-11-10 18:11:40Amani Kiruga, Zhiyi Li, Ruojin Cai, Hansen Lillemark, Xiaoshan Wu, Ayush Tewari, Yilun DuAbstract
Understanding 4D scenes is a core challenge in visual intelligence, requiring models to jointly capture scene geometry, appearance, and dynamics from partial observations. However, existing approaches are subdivided into feedforward models for direct geometric prediction, and generative models for visual synthesis. We introduce UGG, a unified framework for 4D scene understanding that jointly models RGB video and geometric representations such as depth and camera pose. By assigning different noise levels across modalities and time, UGG formulates visual generation and geometric prediction as conditional completion within a single generative model. This flexible formulation enables the model to learn from diverse, heterogeneous datasets with different subsets of geometric and visual annotations. Furthermore, our generative formulation enables inference-time search over multiple plausible scene hypotheses, using geometric evidence extracted from observed RGB frames to improve geometric prediction without ground truth geometry. We demonstrate UGG's effectiveness over non-unified baselines across camera-controlled video generation, depth estimation, and camera pose prediction, including in challenging sparse observation settings. Code and visualizations are available at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2611.14440
PDF
-
Bits or permutations? Spending a fixed inference budget on weight precision or on option-order averaging in small local language models
2026-11-10 17:53:44Mohammed Nayeem Khan, Logesh L, Aliyas Ahmad MisriAbstract
Local language models can save memory by storing weights in fewer bits, and they can reduce sensitivity to the order of answer options by averaging a few reorderings. We ask whether the first makes the second worse, and which is the cheaper way to spend a fixed budget. On Qwen3-0.6B and Qwen3-1.7B with simulated round-to-nearest weight quantization (8, 6, 4 and 3 bits, fp16 reference), read from the logits of the letters A to D on ARC-Easy, ARC-Challenge, MMLU and HellaSwag (300 items each), 6 bits lose no accuracy against fp16 (they score +1.8 (+1.0 to +2.5) and +0.9 (+0.3 to +1.5) points higher, which we do not explain), and 8 bits change it by -0.7 (-1.1 to -0.3) and -0.4 (-0.7 to -0.2) points. At 4 bits the answer-slot skew rises from 0.15 to 0.53 (0.6B) and from 0.07 to 0.12 (1.7B); Gaussian noise at the same weight SNR gives 0.15 and 0.37 (two draws) and 0.19 (one draw), so the control is inconclusive on whether rounding adds slot skew beyond equal-power noise. Flip rates rise but stay below an accuracy-matched baseline of independent uniform errors. Averaging four cyclic orders recovers +12.6 points at 4 bits for 0.6B (+5.3 at fp16). We price bits and passes in one machine-independent unit, weight bytes read times passes. In cost-matched pairs, extra orders beat extra bits only when the weights have 6 or more bits, and the decision table reduces to: use 6-bit weights and spend the rest of the budget on up to four orders, one pass each. Everything is simulated on two small models with the simplest quantizer; a title-level search found no earlier study of quantization together with option order.
Abstract (translated)
URL
https://arxiv.org/abs/2611.14413
PDF
