Upload your solutions at: https://tinyurl.com/AML-2021-ASSIGNMENT1
1. (0.5 points) Give an example of a finite hypothesis class H with VCdim(H) = 2021.
Justify your choice.
2. (0.5 points) Consider Hballs to be the set of all balls in R2
:
Hballs = {B(x,r), x ∈ ℝ2
, r ≥ 0 }, where B(x,r) = {y ∈ ℝ2
| || y – x ||2 ≤ r}
As mentioned in the lecture, we can also view Hballs as the set of indicator functions of
the balls B(x,r) in the plane: Hballs ={ ℎ!.!: ℝ! → 0,1 , ℎ!.! = �!(!,!), � ∈ ℝ!, � > 0}.
Can you give an example of a set A in R2 of size 4 that is shattered by Hballs? Give
such an example or justify why you cannot find a set A of size 4 shattered by Hballs.
3. (1 point) Let X = R2 and consider Hα the set of concepts defined by the area inside a
right triangle ABC with the two catheti AB and AC parallel to the axes (Ox and Oy)
and with AB/AC = α (fixed constant > 0). Consider the realizability assumption. Show
that the class Hα can be (�, �) − PAC learned by giving an algorithm A and
determining an upper bound on the sample complexity mH( �, �) such that the
definition of PAC-learnability is satisfied.
4. (1 point) Consider H to be the class of all centered in origin sphere classifiers in the
3D space. A centered in origin sphere classifier in the 3D space is a classifier hr that
assigns the value 1 to a point if and only if it is inside the sphere with radius r > 0 and
center given by the origin O(0,0,0). Consider the realizability assumption.
a. show that the class H can be (�, �) − PAC learned by giving an algorithm A and
determining an upper bound on the sample complexity mH(�, �) such that the
definition of PAC-learnability is satisfied. (0.5 points)
b. compute VCdim(H). (0.5 points)
5. (1 point) Let H = {ℎ!: ℝ → 0,1 , ℎ! � = � !,!!! ∪[!!!,!!) � , � ∈ ℝ}. Compute
VCdim(H).
6. (1 point) Let X be an instance space and consider H ⊆ {0,1}! a hypothesis space with
finite VC dimension. For each � ∈ X, we consider the function zx: H →{0,1} such
that zx(h) = h(x) for each ℎ ∈ H. Let Z = {zx: H →{0,1}, � ∈ X}. Prove that
VCdim(Z) < 2VCdim(H)+1.
Ex-officio: 0.5 points
n clinical populations, the suicide rate of patients with borderline personality disorder (BPD) is estimated between 8% and 10%, a rate far greater than that in the general population. While suicidal tendencies in BPD patients is a multi-determined phenomenon, a key factor that leads to the higher than average mortality rate could be the impulsive trait common among BPD patients. Hence, a general consensus highlights that by reducing the impulsive personality characteristic, this will ultimately reduce the high risk nature of this disorder. In light of this, Dialectical Behavioural Treatment (DBT) – an adapted version of cognitive behavioural therapy – was developed to specifically reduce suicidality in BPD patients by applying behavioural techniques to control impulsivity. This essay reviewed previously conducted randomised controlled trials, to investigate the efficacy of DBT in treating suicidality in BPD patients through reducing impulsivity, as well as examined the limitations of this intervention. The findings indicated DBT to be more effective in several outcome measures, namely self-harm, parasuicidal behaviours and suicidal ideation, when comparing DBT with treatment as usual and community treatment by experts. Additionally, DBT’s mindfulness module was found to be a salient component in reducing impulsive self-injurious behaviour. While the efficacy of DBT is undisputed, this intervention is still an extensive and costly treatment. Hence, dismantling studies were conducted to identify the treatment component most significant to treat suicidality in BPD and findings revealed shortened variants of DBT are more cost-effective and are just as effective in yielding significant improvements.
INTRODUCTION
Suicide risk is a frequent companion in the treatment of Borderline Personality Disorder (BPD); a disorder that has a prevalence of 1-2% in the general population, and is present in 10% of psychiatric outpatients and 20% of inpatients (Lieb, et al., 2004). According to a study conducted by the Substance Abuse and Mental Health Services (SAMHSA) in 2014, nearly 80% of individuals with BPD report a history of suicide attempts and up to 10% do end up completing suicide – a rate that is almost 50 times higher than in the general public (Sack, 2015). In fact, self-harm and suicide attempts are so prevalent in BPD that it is the only personality disorder to have such behaviours included in its diagnostic criteria. For instance, the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR) includes “recurrent suicidal behaviour, gestures, or threats, or self-mutilating behaviour’’ as the fifth criterion for diagnosing BPD (Oldham, 2006, p. 20). Indeed, there is an increasingly important need to reduce the high mortality rates of this disorder and debunk the notion of BPD as “the suicidal personality disorder” (Pompili, et al., 2005, p. 319). Given the high risk nature of this disorder, it is important to discern wh