These per-frame methods extend the concept oftemporal tracking bymatching frame-wise predictions ofhighsimilarities.

Results ofexperiments showthatRMT performs on par with the Transformer-XL on language modeling for smaller memory sizes and outperforms it for tasks that require longer sequence processing. We show that adding memory tokens to Tr-XL is able to improve its performance.

We propose a novel end-to-end solution for video instance segmentation (VIS) based on transformers. Recently, the per-clip pipeline shows superior performance over per-frame methods leveraging richer information from multiple frames. However, previous per-clip models require heavy computation and memory usage to achieve frame-to-frame communications, limiting practicality.In this work, we propose Inter-frame Communication Transformers (IFC), which significantly reduces the overhead for information-passing between frames by efficiently encoding the context within the input clip.Specifically, we propose to utilize concise memory tokens as a means of conveying information as well as summarizing each frame scene.The features of each frame are enriched and correlated with other frames through exchange of information between the precisely encoded memory tokens.We validate our method on the latest benchmark sets and achieved state-of-the-art performance (AP 42.6 on YouTube-VIS 2019 val set using the offline inference) while having a considerably fast runtime (89.4 FPS). Our method can also be applied to near-online inference for processing a video in real-time with only a small delay.The code is available at https://github.com/sukjunhwang/IFC

Transformer-based models show their effectiveness across multiple domains and tasks. The self-attention allows to combine information from all sequence elements into context-aware representations. However, global and local information has to be stored mostly in the same element-wise representations. Moreover, the length of an input sequence is limited by quadratic computational complexity of self-attention. In this work, we propose and study a memory-augmented segment-level recurrent Transformer (RMT).

Humans and animals show remarkable learning efficiency, adapting to new environments with minimal experience. This capability is not well captured by standard reinforcement learning algorithms that rely on incremental value updates. Rapid adaptation likely depends on episodic memory -- the ability to retrieve specific past experiences to guide decisions in novel contexts. Transformers provide a useful setting for studying these questions because of their ability to learn rapidly in-context and because their key-value architecture resembles episodic memory systems in the brain. We train a transformer to in-context reinforcement learn in a distribution of planning tasks inspired by rodent behavior. We then characterize the learning algorithms that emerge in the model. We first find that representation learning is supported by in-context structure learning and cross-context alignment, where representations are aligned across environments with different sensory stimuli. We next demonstrate that the reinforcement learning strategies developed by the model are not interpretable as standard model-free or model-based planning. Instead, we show that in-context reinforcement learning is supported by caching intermediate computations within the model's memory tokens, which are then accessed at decision time. Overall, we find that memory may serve as a computational resource, storing both raw experience and cached computations to support flexible behavior. Furthermore, the representations developed in the model resemble computations associated with the hippocampal-entorhinal system in the brain, suggesting that our findings may be relevant for natural cognition. Taken together, our work offers a mechanistic hypothesis for the rapid adaptation that underlies in-context learning in artificial and natural settings.

Diffusion Transformer(DiT) based video generation models have recently achieved impressive visual quality and temporal coherence, but they still frequently violate basic physical laws and commonsense dynamics, revealing a lack of explicit world knowledge. In this work, we explore how to equip them with a plug-and-play memory that injects useful world knowledge. Motivated by in-context memory in Transformer-based LLMs, we conduct empirical studies to show that DiT can be steered via interventions on its hidden states, and simple low-pass and high-pass filters in the embedding space naturally disentangle low-level appearance and high-level physical/semantic cues, enabling targeted guidance. Building on these observations, we propose a learnable memory encoder DiT-Mem, composed of stacked 3D CNNs, low-/high-pass filters, and self-attention layers. The encoder maps reference videos into a compact set of memory tokens, which are concatenated as the memory within the DiT self-attention layers. During training, we keep the diffusion backbone frozen, and only optimize the memory encoder. It yields a rather efficient training process on few training parameters (150M) and 10K data samples, and enables plug-and-play usage at inference time. Extensive experiments on state-of-the-art models demonstrate the effectiveness of our method in improving physical rule following and video fidelity. Our code and data are publicly released here: https://thrcle421.github.io/DiT-Mem-Web/.

Large Language Models (LLMs) have been shown to be able to learn different tasks without explicit finetuning when given many input-output examples / demonstrations through In-Context Learning (ICL). Increasing the number of examples, called ``shots'', improves downstream task performance but incurs higher memory and computational costs. In this work, we study an approach to improve the memory and computational efficiency of ICL inference by compressing the many-shot prompts. Given many shots comprising t tokens, our goal is to generate a m soft-token summary, where m < t. We first show that existing prompt compression methods are ineffective for many-shot compression, and simply using fewer shots as a baseline is surprisingly strong. To achieve effective compression, we find that: (a) a stronger compressor model with more trainable parameters is necessary, and (b) compressing many-shot representations at each transformer layer enables more fine-grained compression by providing each layer with its own compressed representation. Based on these insights, we propose MemCom, a layer-wise compression method. We systematically evaluate various compressor models and training approaches across different model sizes (2B and 7B), architectures (Gemma and Mistral), many-shot sequence lengths (3k-6k tokens), and compression ratios (3x to 8x). MemCom outperforms strong baselines across all compression ratios on multiple classification tasks with large label sets. Notably, while baseline performance degrades sharply at higher compression ratios, often by over 20-30%, MemCom maintains high accuracy with minimal degradation, typically dropping by less than 10%.

Web platforms, mobile applications, and connected sensing systems generate multivariate time series with states at multiple levels of granularity, from coarse regimes to fine-grained events. Effective segmentation in these settings requires integrating across granularities while supporting iterative refinement through sparse prompt signals, which provide a compact mechanism for injecting domain knowledge. Yet existing prompting approaches for time series segmentation operate only within local contexts, so the effect of a prompt quickly fades and cannot guide predictions across the entire sequence. To overcome this limitation, we propose MemPromptTSS, a framework for iterative multi-granularity segmentation that introduces persistent prompt memory. A memory encoder transforms prompts and their surrounding subsequences into memory tokens stored in a bank. This persistent memory enables each new prediction to condition not only on local cues but also on all prompts accumulated across iterations, ensuring their influence persists across the entire sequence. Experiments on six datasets covering wearable sensing and industrial monitoring show that MemPromptTSS achieves 23% and 85% accuracy improvements over the best baseline in single- and multi-granularity segmentation under single iteration inference, and provides stronger refinement in iterative inference with average per-iteration gains of 2.66 percentage points compared to 1.19 for PromptTSS. These results highlight the importance of persistent memory for prompt-guided segmentation, establishing MemPromptTSS as a practical and effective framework for real-world applications.

Continuous, high-frame-rate, high-resolution processing of long video streams is critical for future AI agents, yet current video-understanding LLMs struggle to scale. Offline, fixed-frame-number methods require the stream length to adapt frame rates; streaming methods constrain memory by merging or discarding tokens, losing information. We propose video-SALMONN S, a streaming audio-visual LLM that, to our knowledge, is the first to process 3-hour videos at 1 FPS and 360p resolution under a fixed memory budget. Our model introduces (i) a test-time-training (TTT) memory module that continually updates token representations to capture long-range dependencies by replacing token merging, and (ii) a prompt-dependent memory reader that selectively retrieves context-relevant content from fixed-size memory. The TTT module is optimised with a Hessian-free conjugate-gradient procedure (TTT_HF) for efficient adaptation. On long-video benchmarks (Video-MME, LVBench, VideoEvalPro), video-SALMONN S sustains high-quality understanding on multi-hour videos with 10k frames and 1M tokens. Our 8B-parameter model achieves 74.2% overall and 67.8% on the Video-MME long split, outperforming both offline and streaming baselines.

Large language models rely heavily on prompts to specify tasks, recall knowledge and guide reasoning. However, this reliance is inefficient as prompts must be re-read at each step, scale poorly across tasks, and lack mechanisms for modular reuse. We introduce TokMem, a tokenized procedural memory that stores recurring procedures as compact, trainable embeddings. Each memory token encodes both an address to a procedure and a control signal that steers generation, enabling targeted behavior with constant-size overhead. To support continual adaptation, TokMem keeps the backbone model frozen, allowing new procedures to be added without interfering with existing ones. We evaluate TokMem on 1,000 tasks for atomic recall, and on function-calling tasks for compositional recall, where it consistently outperforms retrieval-augmented generation while avoiding repeated context overhead, and fine-tuning with far fewer parameters. These results establish TokMem as a scalable and modular alternative to prompt engineering and fine-tuning, offering an explicit procedural memory for LLMs.