Intelligent reflecting surface (IRS), which is a low-cost, low-complexity and easy-to-implement device, has drawn considerable attention recently both in academia and industry as a promising new solution to meet the high demand for mobile network coverage, ubiquitous access capability, air interface transmission rate, etc., in future wireless networks. Due to the unique hardware properties of IRS (composed of a large number of passive reflecting elements), developing low-cost transmission mechanisms for IRS-enhanced communication systems and optimizing their performance are the main research focuses and difficulties. In order to address these issues, this talk will focus on channel information acquisition, beamforming, and robustness design, and propose efficient IRS-assisted transmission mechanisms tailored for typical IRS communication scenarios. Meanwhile, based on (stochastic) optimization, compressed sensing and deep learning techniques, this talk will design highly-reliable data-model-knowledge hybrid driven joint optimization algorithms with fast convergence. The theories, techniques, and design approaches/algorithms involved in this talk would contribute to the maturity of IRS-aided communication systems, and push the fundamental research results towards real-world applications and large-scale deployment in next generation mobile (or other special) communication systems.