I find myself attracted to Springer's Graduate Texts in Mathematics and buy too many of them. Then they sit somewhere at home so that I can pick them up and read them whenever I get a chance. These are essentially filling in background information that I don't have or don't remember (I'm as CS prof with a rusty major in mathematics). If I find a topic that particularly interests me, I chase it down using http://arxiv.org/ and plain old google. This is more or less my forward approach to learning new things. I also have a backwards approach, consisting of finding a paper that appeals to me and then trying to fill in the backwards knowledge required to understand it. Of course, without doing the exercises or applying the stuff I read, my level of understanding is not as deep as it could be.