With the development of AI, the boundaries of multimedia research continue to expand. Author | Zhang Chi
Editor | Ma Xiaoning
On November 10 this year, ACM Multimedia (ACM MM), a premier international academic conference in the field of multimedia, will be held in Rio de Janeiro, Brazil. Leiphone (official WeChat account: Leiphone) will also be on-site to provide firsthand coverage.Welcome to scan the QR code at the end of the article, join our group, and exchange ideas with us.
Compared with previous years, this year's conference has a particularly significant change: the call-for-papers topics have been reorganized.
At the end of last year, the ACM MM Academic Steering Committee voted to adopt a new framework of research topics. The previous five themes and 15 subareashave been replaced with 16 coequal research tracks.
At first glance, this appears to be merely a reclassification, but comparing the old and new lists reveals that some research has been placed in more prominent positions. After making the comparison, we identified several notable changes.
Adding Embodied Intelligence and Multimodal Reasoning
Most notably, two newly added standalone topics have been added:Embodied and Immersive Multimedia(Embodied and Immersive Multimedia) andMultimedia Reasoning(Multimedia Reasoning).
The official description of the former specifically mentions intelligent agents capable of perceiving, acting, and interacting in the real world, as well as the integration of information across multiple modalities, including vision, language, audio, and touch.
That said, this does not mean that ACM MM is turning into a robotics conference; it emphasizes multimedia and multimodal research in embodied intelligence, rather than robot hardware, control, or motion planning.
Traditional multimedia research primarily deals with content such as images, sound, and video, while embodied systems need to connect multimodal perception with action.
Multimedia Reasoningemphasizes logical, causal, temporal, and spatial reasoning over images, video, speech, and text. For example, a model must not only recognize what happened in a video but also determine why the event occurred, how different clues relate to one another, and what information is not directly conveyed by the visuals.
Multimedia AI in the past placed greater emphasis on extracting information from content; the conference now presumably seeks to highlight research on reasoning based on multimodal evidence as a distinct area.
However, “understanding” and “reasoning” are not completely separate either; past research has also involved reasoning.
02From Vision and Language to Multimedia and Language
Another change is that the 2025 “Vision and Language” (Vision and Language) was changed to “Multimedia and Language” (Multimedia and Language). The new description explicitly incorporates audio and sensor data in combination with language, alongside images and video.
Meanwhile, the previously separate “Multimedia Foundation Models” and “Generative Multimedia” were merged in 2026 into “Multimedia Generation and Foundation Models”. The conference no longer treats foundation models and content generation as entirely separate discussion topics, but instead emphasizes their connections in model architectures, cross-modal alignment, and controllable generation.
These developments all reflect an expansion and integration of the kinds of information that multimedia processes.
03From Discussing Impact to Taking Responsibility
A 2025 topic called “Social Impact of Generative AI” (Social Aspects of Generative AI) was changed in 2026 to “Responsible Multimedia” (Responsible Multimedia).
From emphasizing social impact to emphasizing responsibility, the intent is clear: AI’s impact is already real and tangible, and what needs to be considered now is how to control it. Just as there were discussions two years ago about whether AI-generated fake videos would mislead the public, the focus now should be on how to eliminate such misinformation.
In addition, “Multimedia Interpretation” (Multimedia Interpretation) was adjusted to “Multimedia Interpretability and Explainability” (Multimedia Interpretability and Explainability).
Interpretability and Explainability have similar meanings, and there is no uniform standard for distinguishing them. In general, the former focuses more on whether the internal working mechanisms of a model can be understood, whereas the latter emphasizes whether the basis for a model’s judgment can be explained.
This actually reflects the fact that AI is increasingly involved in judgment and decision-making, and even conducting research on its own (such as the recently much-discussed RSI recursive self-improvement); merely having it provide results is no longer enough. We also need to know where those judgments come from and whether they are reliable.
Has Multimedia Changed?
ACM MM was founded in 1993.
At that time, multimedia research was still focused on relatively basic questions: how to store, transmit, and synchronize images, sound, and video on computers, and how to retrieve and organize such content.
In 2004, the conference divided research into three major areas: content, systems, and applications. Later, as the Internet, mobile devices, and deep learning rose in turn, research in areas such as multimedia retrieval, content understanding, and human–computer interaction also continued to evolve, and the conference’s research scope expanded accordingly.
In recent years, this change has accelerated further. In 2024, ACM MM added the dedicated theme “Multimedia in the Era of Generative AI”; this year, directions such as embodied intelligence and multimodal reasoning were separately listed once more.
Over these thirty-plus years, the core of ACM MM research has been how to process and use information in different forms. With the arrival of AI, the research questions have also evolved further: from storing, transmitting, and retrieving multimedia content, to understanding and generating content, and then to enabling machines to reason and make judgments based on multimodal information, and even interact with the real environment.
The 2026 adjustment does not actually mean that traditional multimedia research is leaving the stage; rather, as AI continues to develop, the research boundaries of multimedia are further expanding.
References:
ACM Multimedia 2025 Topics of Interest:https://acmmm2025.org/research-topics/
ACM Multimedia 2026 Topics of Interest:https://2026.acmmm.org/site/topics.html
