Covers mid- and high-level vision: object detection and tracking, semantic/instance segmentation, open-vocabulary and few-shot recognition, 3D vision (point clouds, depth, Gaussian splatting, reconstruction, pose, SLAM), human pose and action recognition, autonomous driving and LiDAR perception, remote sensing and hyperspectral/SAR image analysis, face and person re-identification, anomaly detection in images, and video understanding.
Covers models that connect vision, language and other modalities: multimodal large language models and vision-language models (VQA, captioning, grounding, hallucination, reasoning), CLIP-style contrastive alignment and prompt learning, cross-modal retrieval and matching, audio-visual learning and fusion, multimodal fusion with missing modalities, multimodal sentiment and sarcasm, video-language understanding, and multimodal agents.
Focuses on text-centric large language models: reasoning and chain-of-thought, reinforcement learning post-training and preference alignment, instruction tuning and parameter-efficient fine-tuning (LoRA), retrieval-augmented generation and knowledge, LLM agents and tool use, hallucination and factuality, efficient LLM inference (quantization, KV cache, speculative decoding), LLM safety and jailbreaks, and NLP tasks such as text classification, information extraction, translation, summarization and recommendation with LLMs.
Covers making models cheaper and distributed: network pruning and sparsity, quantization and low-bit inference, knowledge distillation, low-rank and tensor compression of networks, efficient architectures (MoE, early exit, dynamic networks, spiking neural networks), neural architecture search, hardware-aware and on-device deployment, and federated, decentralized and communication-efficient learning.
Covers learning under limited, shifting or noisy supervision: domain adaptation and generalization, test-time adaptation, continual and incremental learning, few-shot and zero-shot learning, meta-learning, self-supervised and contrastive representation learning, semi- and weakly-supervised learning, learning with noisy labels and class imbalance, out-of-distribution detection and open-set recognition, active learning, and calibration and uncertainty.
Addresses trust and protection of signals and models: adversarial attacks and defenses on vision, speech and multimodal models, backdoor and poisoning attacks, deepfake and AI-generated image/video/audio detection, image forgery localization and media forensics, watermarking and copyright protection (images, audio, 3D, models), privacy (differential privacy, membership inference, gradient leakage, anonymization), machine unlearning, steganography and steganalysis, and physical-layer and biometric security.
Covers signal processing for communication and RF sensing: MIMO and massive MIMO, channel estimation and prediction, beamforming and precoding, reconfigurable intelligent surfaces, OFDM/OTFS waveforms, integrated sensing and communication (ISAC), semantic communications, radar signal processing and target detection, direction-of-arrival estimation, localization and positioning, and AI-native 6G air interfaces.
Includes low-level visual signal processing: image and video super-resolution, denoising, deblurring, deraining/dehazing and all-in-one restoration, low-light and HDR enhancement, image and video coding (learned and standard codecs, VVC), point-cloud and neural-field compression, image signal processing and demosaicing, image/video quality assessment, frame interpolation, watermarking-free low-level vision, and computational imaging and inverse problems for photography.
Develops the mathematical core: estimation and detection theory, Bayesian inference and particle/Kalman filtering, sparse recovery and compressed sensing, tensor and low-rank methods, convex and non-convex optimization, distributed and stochastic optimization, adaptive filtering, information theory, random matrix theory, quantum signal processing, and theoretical analysis of learning algorithms.
Covers image-based medicine: medical image segmentation and classification (MRI, CT, X-ray, ultrasound, endoscopy, dermoscopy, retinal), pathology and whole-slide imaging, MRI/CT/PET reconstruction and acceleration, medical image registration and synthesis, medical vision-language models and report generation, surgical video analysis, and federated or privacy-preserving medical imaging.
Focuses on generating visual and general content: diffusion and flow-matching models and their theory, GANs and VAEs, text-to-image and text-to-video generation, image and video editing, controllable and personalized generation, talking heads and avatar animation, 3D and motion generation, layout/graphics generation, generative data augmentation, and acceleration or safety of generative models (excluding speech/audio generation, which has its own topic).
Covers converting and understanding speech: automatic speech recognition and speech foundation models (Whisper-style, SSL encoders), speech LLMs and spoken language understanding, multilingual and low-resource ASR, streaming and contextual recognition, keyword spotting, audio-visual and silent speech recognition, spoken dialogue and speech translation.
Covers general audio and music: sound event detection and audio tagging, acoustic scene classification, audio captioning and audio-language models, text-to-audio and music generation, music information retrieval (transcription, beat/chord/structure, source separation of music), audio foundation models and representation learning, audio retrieval, and AI-generated music detection.
Focuses on temporally ordered data: time-series forecasting (long-term, multivariate, probabilistic), time-series foundation models and LLMs for time series, anomaly detection in sensor, log and industrial data, time-series classification and representation learning, imputation and irregular sampling, traffic and energy forecasting, financial time series, predictive maintenance and fault diagnosis, and dynamical-systems and physics-informed modeling.
Focuses on signals and learning on graphs: graph signal processing (graph filters, graph Fourier transform, sampling and recovery on graphs, graph learning from data), graph neural networks and their theory (expressivity, oversmoothing, heterophily), dynamic and temporal graphs, knowledge graphs, hypergraphs and simplicial complexes, graph clustering and community detection, and applications such as recommendation and anomaly detection on graphs.
Applies signal processing to physiological data: EEG and brain-computer interfaces, ECG/PPG and cardiovascular monitoring, fetal and neonatal monitoring, sleep staging, wearable and IMU-based activity and gait analysis, EMG and neural decoding, respiratory and acoustic health sensing, remote photoplethysmography, and clinical decision support from time-series physiology.
Focuses on who is speaking and how: speaker verification, identification and diarization, speech emotion recognition and multimodal emotion/sentiment analysis, speaker and language identification, age/gender/health-related paralinguistics from voice, depression and cognitive-decline detection from speech, and anti-spoofing/deepfake speech detection.
Includes cleaning and separating audio signals: speech enhancement and noise suppression, dereverberation, speech separation and target speaker extraction, acoustic echo cancellation, bandwidth extension and packet-loss concealment, generative (diffusion) enhancement, hearing aids and intelligibility prediction, and perceptual/objective quality metrics.
Includes the physics and geometry of sound: sound source localization and direction-of-arrival estimation for audio, microphone-array beamforming, room acoustics and impulse-response estimation, sound field reconstruction and spatial audio rendering (binaural, ambisonics, HRTF), acoustic sensing and ultrasound, underwater acoustics, and active noise control.
Focuses on generating speech: text-to-speech and zero-shot voice cloning, voice conversion and speaker anonymization, singing and expressive/emotional synthesis, prosody and duration modeling, codec- and diffusion/flow-based speech generation, vocoders, speech editing, and evaluation of synthesized speech.